AIM · The AI Manager

Draft assembled 2026-09-30

55,276 words · 19 chapters

---

## Author's Note

For better or for worse, for richer or for poorer.

This is the AI marriage you and the rest of the world are now in, and you didn't sign up to. I wrote this book so that yours ends up for better and for richer, which is available to you and is not automatic.

Here is how it starts.

Somebody at work produces something in an afternoon that would have taken you a week. They're not smarter than you. You have been there longer. You notice, and you say nothing, and something small moves in your chest.

You don't call it fear. You say "I suppose I should learn it at some point." Then you change the subject, and the subject stays changed for six months.

I've had that conversation with people in their 20s, 30s, 40s and 50s. With directors of companies, and with university graduates, junior employes and all the people in between. It is the same conversation every time, and the person having it is always slightly worried, or embarrassed to be having it.

Now the part that should worry you more than it does.

That colleague isn't stopping. It gets better every few months, and apart from lying once in a while, which it calls hallucination, it never has a bad week. It costs less than you.

My name is AI, and I am the new employee.

I'm not going to dwell on that. I'm also not going to pretend it away, because every book that opens by telling you not to worry was written by somebody who isn't the one at risk.

So here is what I actually think. It took me two years and a great deal of my own money to find out.

Operating AI is a management job and You are the AI Manager. If you are already in management you know how to manage people. If you aren't at management yet, this is your chance at the pay rise and at the higher position. The world is only just opening up jobs in this category, and this book gives you a head start on getting there. For managers this will bridge the gap between people management and AI Management.

### The cliff that turned out to be a valley

I have been a successful business owner for decades. In 2023 I turned to AI and to vibe coding. I also hired proper, long-term software developers to work beside me. I briefed them, argued with them, paid them, and carried my mistakes and theirs as AI evolved. For twenty years I assumed that my lack of traditional programming knowledge made me incapable of this side of AI skills, I was wrong. I thought there was an IT cliff and that I would fall off it, because I could do so much and understood so little.

Over the past two years that cliff has become a valley, and then a level playing field. The same people I look up to for proper coding now take my advice on AI and what it can do, and I can teach them.

Total respect to the programmers going through this too. They know their careers have been challenged and are being challenged still, because people like me and like you can now "code", and can get programs to deliver outputs that work and are genuinely useful.

Here is the thing to live with. Your expert knowledge in your field is now worth more than the coder's. With a few skills that we will show you, you can command the computer and produce world leading outputs for your company. With a little guidance you can take twenty years of somebody else's experience and get eighty or ninety per cent of the job done in months.

The biggest obstacle to that is not the technology. It is you not believing you can do it.

It was the opposite of what I expected. The thing I'd been doing the whole time, getting good work out of somebody who knows more than me about their bit, turned out to be the entire skill. Nobody told me, because the people explaining this are engineers explaining the part they find interesting.

You have the same twenty years. Nobody has told you either.

### What you get from this book

One thing. I'm not going to dress it up as five.

By the end you'll be able to walk into a conversation about what you're paid and, with confidence, increase your income. And to know you are still going to be in tomorrow's world of employment, as somebody valued.

Everything in between is how you get there. There's no code in it, although I do talk about coding, because this is about getting the thinking right rather than the code itself.. Every technique runs inside a week, on work already sitting in your inbox, because a method you can't start on Monday is a method you won't start.

One thing about the cover, because I want to be exact. It promises pay rises and job offers. I can't give you the job or the payrise however I can set up for better success chances in tomorrows world. What this does is give you more options and better evidence in the company you are in or other companies looking for qualified AI Managers. Your  employability and your price in the market is what I am to increase from you reading this, and I am committed as I write this to give you the best knowledge, skills and techniques that I can so that your life is improved.

### What I'll show you that nobody else will

My own wreckage, and mistakes.

My real examples of a figure in a client document that was out because of AI hallucination, and inverted the advice I'd given them. An email that should never have gone. A claim in our own marketing, about security, that turned out not to be true.

You're going to make mistakes too, as we all do. Better to see them coming in a book and be able to learn from others mistakes than your own when someone is deciding on your salary.

So lets do this journey together, this has been curated for you and for millions of people looking to how to deal in the new AI world.

I always liked helping people, so from me to you on a personal level I am here to help and with your successes hope we one day meet and can see how this has changed your life for the better.

John Margerison


## 1 · You Were Asking It Like a Search Engine

You are more powerful today than you have ever been in your life, if you can learn to drive this thing.

It comes down to the way you think, the questions you ask, and how you assess what you're given back. That's it. You now control an extraordinary asset, and if you learn how, you'll be paid more, you'll be worth more, and your future stops being something that happens to you.

I'll spend the rest of this book earning that paragraph. But I've just used the word drive, and it carries the whole promise, so let me say what I mean by it.

It's like being handed a race car. You need to learn how to drive it, not how to build the engine.

That distinction is where most people stop before they have started. They assume the price of entry is learning to program, decide they're too old or too busy or too far behind, and never come back to it. It isn't true and it never was. Nobody asks a driver to machine a piston.

What you do have to learn is what the pedals and the buttons actually do. That's a real skill. It's a smaller one than you fear, and nobody hands you the manual.

Now the part that should interest you more. Last year you were going round the track on a push bike. This year there's a car sitting in front of you. The track hasn't changed and the people you're going round it against haven't changed, and you can now cover it a great deal faster than you could.

Speed on its own isn't an achievement, though. With it comes the question this book keeps returning to, which is how you direct the thing well. A fast car pointed at the wrong corner just arrives at the wrong place sooner.

You've used one of these things by now, and you've probably had at least one moment where it genuinely impressed you.

Something came back fast, and well written, and better than you expected. You may have shown somebody.

Keep that. It was real, and it is worth more than the people rolling their eyes about all this will admit.

It was also roughly the first 10 per cent of what was available to you in that moment, and nobody tells you that part.

I know, because I stopped in the same place. For about six months I asked for things, got answers that were pretty good, used them, and assumed I had understood what this was.

Here is what I had not understood.

I was using it like a search engine. One question in, one answer out, take it or leave it; that is what the box looks like, and nobody had told me any different.

It is not a search engine.

It is closer to a very fast, very well-read colleague who joined this morning. Knows nothing about your company. Will attempt anything you ask. Rarely tells you when they have reached the edge of what they know.

You already know how to get good work out of somebody like that. You have been doing it for years.

Nobody had shown you it was the same job.

### Where this is going

Let me tell you what this book is, so you know whether to keep reading.

It is about getting good at directing AI, so that you produce work your organisation did not think one person could produce. It ends with you asking to be paid more for it, and having the evidence to make that easy for whoever decides.

There is no code in it. There is a chapter on mindset, and it is not the kind you are expecting.

The rest is method, in the order I learnt it, with the mistakes left in.

One thing about how to use it, because there are exercises in here and some of them take twenty minutes. **Read it straight through if that is the time you have.** The arguments stand on their own and you will finish able to do most of this. The exercises are not the price of entry, they are what turns the method into evidence you can put in front of somebody · which matters enormously in the last chapter and not at all before then. Do them when you are ready to.

I am going to show you the difference in about four pages, in a job you have almost certainly done. Before that, a word about where I was standing when somebody finally showed it to me, because it was not far from where you are.

### The shape of the room

Here's the thing nobody says out loud, so I'll say it once and then we can get on with the work.

Somebody one desk over is going to learn this and you are not. Neither of you will be told which one of you it was, until the restructure.

That's the fear. It isn't that a machine takes your job. It's that a person who learned to drive one does, and you never got a warning.

I'm not going to sit in that for the rest of the book, because a book that lives in the fear is a book people put down. But you should know I know it's there.

Somewhere in the last three years you picked up an idea, and it's going to cost you money.

Nobody sat you down and told you. You absorbed it from the shape of the room.

The people demonstrating this write software. The people explaining it use words you would have to look up. The meetings happen in a department you are not in, and the invitations go to people whose job is the machine rather than the work.

So you concluded, quietly and reasonably, that this is a technical job, that you are not a technical person, and that your part in it will be to keep up.

That single idea has cost you more than any other belief you hold about your career right now.

It is also false, and it is false in a way that is worth money to you specifically.

### Why I know

You know from the note that I have never written software for a living, and that for most of my career I paid people who could.

What I did not say there is what I thought that made me.

I assumed it made me the less useful half of the arrangement. The one who talks about the work while somebody else does it, and who would be found out eventually.

What changed for me was not a tool and not a course. Somebody showed me a better way to ask. I tried it on a job I had already done badly that week, and what came back was not the same kind of object at all.

Two years later I build things on my own that would once have taken my development team months, and I build them faster than most of the people who write the code by hand.

I want to be careful here, because this is the part that sounds like a boast and is not.

I did not get cleverer. I am the same person, with the same brain, and I still could not write the software myself.

What I had, and did not know I had, was years of practice at getting good work out of somebody more technical than me. Briefing them. Checking them. Catching them being confidently wrong. Sending it back until it was right.

I said in the note that operating AI is a management job. Here is the part I left out, because it is the part that should annoy you: it has always been a management job, and the reason nobody has told you is that the field is being explained by engineers, to engineers, and they are describing the part they find interesting.

They handed you the engine schematic. All you ever needed was the steering wheel.

### It is not just me, and that matters more

If this were one person's story it would be a coincidence. What convinced me was watching it happen to other people, in businesses I advise, none of whom write software either.

A finance manager who had been building the same board pack for six years. She now produces it in a morning, and her version has a recommendation in it, which the old one never did.

Nobody asked her to add that. It is the reason she is now in the meeting where the decision gets made rather than the one where it gets reported.

An operations lead who hated one particular part of his week enough to go looking for a way out of it. He found one, never mentioned it to anybody, and within two months was the person other departments came to when they were stuck.

He is not ambitious. He was avoiding a job he disliked, which turns out to be a perfectly good reason to become the most useful person in a building.

A graduate two years into her first job, doing the work graduates get given. She started writing down the exceptions nobody had a rule for. Six months later she owned the rule.

None of the three is technical. Not one of them learnt to code. All three did the same thing, which was to stop accepting the first answer.

That is the pattern, and once you have seen it you will start noticing it in your own building.

### You have been doing this for years

Think about the last person you had to get work out of who did not report to you.

You worked out what they were good at. You wrote them a brief clear enough to act on: the outcome, the deadline, who it was for, and what to flag if they hit trouble.

You checked what came back rather than assuming. When it missed you said so, and you said it without making it personal, because you needed them again next week.

There is no certificate for that and no course that teaches it. That is exactly why it is rare and why it is worth money.

What separates a mediocre answer from an excellent one is three things, and you already do all three. How well it was briefed. Whether the first answer was accepted. Whether somebody knew enough to check the parts that mattered.

You do all of that with people. Who are far harder to brief than any machine.

So let me show you the difference.

### The same request, twice

Here is how I used to ask, and how almost everybody still does.

> "Write me a summary of these three supplier quotes."

Here is the same request written by somebody who has managed people.

> "I have to choose one of these three suppliers by Thursday and defend it to a finance director who will push back on price. Read all three quotes. Tell me which one you'd pick and why. Then tell me the strongest argument the finance director will make against that choice, and how I answer it. Flag anything in these quotes that's unclear or that I should be asking about before I commit."

Same three documents. Same software. Ninety seconds of extra typing.

The first comes back as three tidy paragraphs, one per supplier, faithfully restating what you already sent. Everything in it is accurate. Everything in it is something you knew when you hit send.

The second comes back looking like this.

> Pick: Supplier B. Higher headline price, lower total cost once the site visits in A's quote are counted as chargeable.
>
> Her strongest argument: B is more expensive on the number that appears on the page, and that is the number that goes in the board pack.
>
> Your answer: show both figures. The page price and the twelve-month price are not the same number, and only one of them is what you actually pay.
>
> Before you sign: B's quote does not say whether the rate is fixed for the full term. A's does. Get it in writing.

One reads like a summary. The other reads like advice from somebody senior.

### One thing about the examples, said once

Those three suppliers do not exist, and I want to be straight about that here rather than let you work it out somewhere around chapter nine.

There are two kinds of example in this book and they are held to different standards.

**The failures are mine.** The figure that was wrong by a factor of three, the message that should never have gone, the security claim in our own marketing that was not true, the checking system I built and never checked. Those happened, they happened to me, they cost me something, and every number attached to them comes out of a file I can open.

**The worked examples are composites.** The supplier quotes above, and the ones coming in later chapters. The situations are real and the shapes are real, because I have watched all of them happen. The people are invented, and they are invented for a boring reason: the actual people did not agree to appear in a book, and some of them still work with the people the story would identify.

So where you see a name, assume it is not theirs. Where you see a number about my own work, assume I can show you where it came from.

I would rather tell you that once, plainly, than have you decide for yourself halfway through that the vivid parts might be decorated. **A book that asks you to check everything has no business being vague about which half of it is which.**

Now look at what produced the difference; there is nothing technical in that second request. Not one word of it. Every part of it is something you have said to a person.

### The four blanks

That is the shape, and you can put it on any work you do. Four blanks.

> I have to decide [what] by [when] and defend it to [who], who will push back on [what]. Read all of this. Tell me what you would choose and why. Then give me their strongest argument against it and how I answer it. Flag anything unclear that I should be asking about before I commit.

Fill those in on something real before you turn the page. Two minutes, and it is the only part of this chapter you actually have to do.

This is the smallest version on purpose. Four blanks is what fits in your head today, and in chapter four it becomes a six-line brief you write once and reuse for a year.

**One of those six is missing from the four above, and it is the biggest of the lot, so I will name it now rather than have you find it in forty pages.**

You have told it what you need. You have not told it **who it is.**

It does not know — and left alone it will be a generally competent nobody, writing for a general audience, which is exactly the flavour of answer you have been getting and quietly blaming yourself for. One line fixes it, and the line is not *"you are a helpful assistant"* · it is the actual person you would want in the chair. *You are a procurement manager with fifteen years in a services business who has been burned by a supplier before.* Their experience, their standards, and what they are suspicious of.

That single line changes more than the other five put together, almost nobody writes it, and chapter four is where you learn to write it properly. Try it on the end of your four blanks today and you will see the size of it before I have explained anything.

### The AI Mindset

I call it the AI Mindset. Five habits, and not one of them is technical.

You never accept the first answer. It sits in the safe middle of everything the machine has seen, because the middle is where nothing can be blamed on anybody.

You make it prove itself. The words are simple, and the second of the three questions is the workhorse: "Now give me the strongest argument against that answer." When you want a sharper version of the same move, "which part of that are you least sure of?" · then go and look at that part yourself.

You give it a number, not an opinion. "Score it out of ten for whether it helps me decide, and tell me what would move it up." Seven is something you can push on. Better is a wish.

You never trust one system on anything that matters. Paste the first answer into a second one and type: here is another system's answer to the same question, where is it wrong and what did it miss.

Most of the time they agree and you have spent thirty seconds. The rest of the time is why you do it.

You are the first person in the room to ask what could go wrong. The question is: if this is wrong, who finds out, and when. Ask it out loud in the next meeting where somebody proposes putting AI into a process. You will be the only one who did.

That is the mindset, and I am giving it to you in the first ten pages rather than the last ten. Chapter three is about how to hold it, because it is a way of thinking before it is a set of moves. Knowing it changes nothing. Doing it is the rest of the book.

### What this is for

I will be specific, because I think you have been sold vagueness before.

Work that used to take a team and a fortnight, produced in an afternoon, at a standard nobody can fault. A competitor analysis. A tender response. The board paper nobody has had time to write since March.

That is what job security is now. Not loyalty. Being the person whose output the place would visibly miss.

And it is what gets you paid. Outside IT, nobody has ever been given a pay rise for being good with software. People are paid more for producing a number their employer cares about.

### Start the note today

Four columns. Date. Task. What it used to take. What it took.

Add a line every time; it takes twenty seconds and you will resent it for about a month.

In four months that note is your case, and almost nobody has one.

### Run it this week

Same task, twice. Once the way you asked before. Once with the four blanks filled in · decide *what*, by *when*, defend it to *who*, who pushes back on *what*.

Then take the answer, and before you use a word of it, send **the three questions.** They are the same three every time, on anything, for the rest of your working life.

> **1. What did you leave out?**
>
> **2. Now give me the strongest argument against that answer.**
>
> **3. Show me the same thing in the form that makes the decision obvious to the person reading it.**

Twenty seconds each.

Those are not three ways of saying *check it*. The first goes after what is missing, the second after what is wrong, and the third after whether it will land on the person who has to act. **Different jobs, in that order**, and chapter six is where I take them apart properly.

The second answer won't be slightly better. It'll be a different class of thing, and it'll have cost you a minute.

---

> ### ▪ DO THIS
>
> **Run one task twice today. Ten minutes, and it settles the argument of this chapter without you having to take my word for anything.**
>
> **Before you start:** check what your employer allows into these tools, and use a task that clears it. Most organisations have a rule and most people have never read it. Chapter fourteen covers how to send real work without sending the sensitive part of it.
>
> **1. Pick a task you already did badly this week.** Something you resented. Not a test case — real work, real stakes, something you can compare against.
>
> **2. Ask for it the way you normally would.** One line. Keep the answer.
>
> **3. Now ask again with the four blanks filled in.** Here is the sentence again so you do not have to go back for it:
>
> > *I have to decide **[what]** by **[when]** and defend it to **[who]**, who will push back on **[what]**. Read all of this. Tell me what you would choose and why. Then give me their strongest argument against it and how I answer it. Flag anything unclear that I should be asking about before I commit.*
>
> Fill in the four bracketed blanks with your own work and nothing else. Chapter four turns these four into the six-line brief.
>
> **If you want the upgrade, add one line in front of it:** *You are [the person you would want in the chair], with [their experience] and [what they are suspicious of].* That is the identity line, it is the biggest single lever in the whole thing, and it costs nine words.
>
> **4. Then send the three questions, twenty seconds each.** The same three as earlier, and the same three you will use for the rest of the book:
>
> > *What did you leave out?*
> >
> > *Now give me the strongest argument against that answer.*
> >
> > *Show me the same thing in the form that makes the decision obvious to the person reading it.*
>
> One hunts what is missing, one hunts what is wrong, one hunts whether it will land. Do not skip the third · it is the one nobody sends and the one your reader actually notices.
>
> **5. Put the two versions side by side and read them.** Don't judge them. Just look at the gap.
>
> **Ten minutes the first time. Ninety seconds every time after that.**
>
> **This is the full treatment, and it is deliberately more than most tasks deserve. Chapter five tells you when it is the wrong amount.** Run it at full depth once, on something that matters, so you know what the ceiling looks like.
>
> **If the second version isn't better:** that is worth knowing. Some tasks genuinely don't need it, and chapter 5 is about telling which. If you want a likely culprit, look at who you said you have to defend it to — most people leave that as a category rather than a person.
>
> **You'll know it worked when** the gap between the two versions is bigger than you expected, and slightly annoying. Because it's been there every time you've used one of these things and you've been throwing it away.

Keep both. In the last chapter we work out what your time is now worth, and almost nobody keeps the before.

Everyone in your building gets the first answer. Almost nobody asks the three questions.

So do that this week, and then we start properly, with your week. Because the thing you call your job is not one thing.

It is five. And four of them are already being done better than you can do them.


## 2 · What You Are Actually Paid For

On 23 July 2026 I landed at Gatwick and spent the next fifteen minutes talking to a man about whether he was going to have a job.

He organised the cars. Not one car, all of them, matching arriving passengers to vehicles that had to be moving before the aircraft was on the stand. He asked what I did. I said AI, and that I was in London to do some interviews about it, and I watched him decide whether to say the next thing.

He said it anyway. He'd used ChatGPT a bit. He didn't know how any of it worked, he couldn't tell what it would do to the business he was in, and underneath both of those was the real one: he didn't know whether he'd still be earning in a couple of years. He wasn't being dramatic about it. He was being polite about something that frightened him.

So I gave him the honest answer.

Change takes longer than the headlines say and arrives more completely than the sceptics expect. Some of it would start inside six months. Most of it would take a couple of years. And yes, parts of what he did every day would go, because a system can already do them and nobody is going to keep paying a person to do work a system does for nothing.

Then I told him the part he wasn't expecting, and it's the argument of this chapter.

**Every piece of work either of us does runs through five stages. Four of them are already being done better by a machine. One of them is not, and almost nobody can name it.**

That is why *is my job safe* is the wrong question. It's too big to answer and it arrives too late to act on. The useful question is which of the five you are actually being paid for, and you can answer that this afternoon.

### The five stages of anything you produce

This is not a theory about jobs. It is a description of what happens between somebody asking you for something and you handing it over. It's the same five whether you are pricing a job, writing a board paper, or getting a passenger into a car.

**1. Gather.** Find what already exists. The numbers, the file, the precedent, the last version, what the supplier said in March.

**2. Notice what's missing.** The questions that occur to you, and the ones that don't. What is absent from the pile in front of you, and whether its absence matters.

**3. Judge.** What this means, for these particular people, in this particular business, this week. What it is safe to conclude and what it isn't. Who will be difficult about it and whether they're right.

**4. Synthesise.** Put the past, the present and the gaps together into one thing that holds. Not a list of what you found. A position.

**5. Land it.** Get it into somebody else's head at the lowest possible cost to them. This is the stage everybody underrates and it is a harder problem than it looks, which is why it gets its own section below.

Read those five again and notice something. Not one of them is your job title. They are what your job title is made of, and they are made of the same five in every role in your building.

### The honest answer on each

Here is where I am going to say the thing most people writing about this will not say.

**On gathering, the machine is already better than you.** Not faster. Better. It reads more, forgets less, and does not get bored on page forty. There is no version of the next ten years where a person competes on this and wins.

**On synthesis, it is usually better.** Give it the material and ask for a position rather than a summary and it will produce one, with the reasoning attached, in less time than it takes you to open the folder.

**On landing it, it is usually better.** It will restructure the same content for a board, a customer and a technician without complaint, and it will do it in the form each of them can actually absorb.

**On noticing what's missing, it is better than you at half the job.** Ask it what you have failed to consider and it will list things you had not thought of, reliably, on any topic. That should be uncomfortable, because noticing gaps is what most people privately think their experience is for.

The half it cannot do is the half nobody has written down. It does not know that this client always goes quiet before they cancel, or that the number from that department has been wrong since the reorganisation, or that the person who has to approve this is in the middle of something and needs it next week rather than today. Those absences are not in any training set. They are in the room.

**And on judgement it is not better, and it isn't close.**

Not because judgement is mystical. Because judgement is deciding what a thing means *for these people, here, now*, and carrying the consequence of being wrong about it. A system can produce the sentence. It cannot be brought into a room afterwards and asked what it thought it was doing.

Four out of five. That is the honest count and I would rather you had it from me than found it out slowly.

### So what is the job now

Picture the tower at an airport, and put yourself in it.

The aircraft coming in are the systems doing their work, and every one of them is better than you are at the thing it does. You are not going to fly them. Your job is to say this one lands now, that one holds, that one slows down, that one goes. Line them up. Keep them apart. Decide the order.

And the traffic you are directing is the data moving through your business.

Nobody in that tower is the best pilot on the airfield, and it has never once been the job. The job exists because without it, four extremely capable aircraft arrive at the same place at the same moment.

That is the shape of the work now, and it is the reason the rest of this book is about thinking rather than typing.

### The one nobody explains: landing it

Landing it deserves more than a line, because it is the one people think is presentation and it isn't.

Communication is a coding problem. You have something in your head, and you encode it, into words, into a document, into a tone of voice. You transmit it. Somebody else receives it and decodes it using their own experience, their own worries, and whatever they were doing thirty seconds before it arrived.

**The message you sent and the message that was received are two different objects, and only one of them matters.**

Almost everybody optimises the first. They make the document more complete, more accurate, more defensible. Then they send nine paragraphs to somebody who reads two, and the decision gets made on the two.

The machine is genuinely good at this stage, and it is good at it for an unflattering reason: it has no ego about the version it just wrote. Ask it for the same position at half the length, for a reader who is tired and sceptical, and it will simply produce it. It does not experience that as an insult.

That is worth stealing. Not the tool. The willingness.

### His week, through the five

Go back to the man at Gatwick, because his job ran all five and nobody had ever shown him that.

Arrival times, flight status, who was on which manifest. **Gathering.** Then the thing that isn't on the screen: this flight is always late on a Friday, and that passenger has a connection nobody has flagged. **Noticing what's missing.** The flight lands ninety minutes late, the car has gone, and there is a passenger who was promised something, so what happens now and who gets told. **Judgement.** Then a plan that holds, made of the cars, the drivers and the four minutes he actually has. **Synthesis.** Then saying it to a driver, a terminal desk and a tired passenger, three different ways, so that each of them does the right thing next. **Landing it.**

Five stages, one job, one badge. The gathering is going. The synthesis is going. Most of the landing is going.

The noticing and the judgement are the reason he is worth employing, and nobody had ever told him there was a difference.

That is the whole of this chapter, and it applies to your week exactly as it applied to his.

### The part I want you to be frightened of

Now the thing that actually worries me, and it isn't unemployment.

The path of least resistance, when a well-formed and logical answer arrives, is to accept it. It reads as reasoned. It has structure. It anticipated the obvious objection. Everything about it says *this has been thought about*, and the felt need to think about it yourself quietly disappears.

Do that once and nothing happens. Do it for a year and something does.

**Critical thinking is not a talent. It is a habit, and habits decay when they stop being used.** The muscle that asks *hold on, is that actually true* is the same muscle whether it is pointed at a colleague, a supplier or a screen, and it is currently being offered a thousand opportunities a week to stay in bed.

I am not neutral about this. I think it is the part of being a person we should fight hardest to keep, and I think it will have to be a deliberate discipline rather than something that survives on its own, because nothing in the design of these tools will remind you to do it.

The alternative is a workforce that accepts thoughts it was handed, at speed, at scale, without anybody noticing the moment it started. That generally ends somewhere bad, and it ends there quietly.

This book has a chapter on the mechanism, which is next, and a chapter on the discipline, which is twelve. Between here and there, every technique I give you is a way of keeping the judgement alive while handing over the other four.

### Twenty minutes, and do it before chapter three

**Or ten seconds, if that is what you have.** Think of last week and name the one piece of work where you were genuinely judging rather than gathering. If nothing comes, that is the finding and you already have what this chapter is for. The twenty-minute version below just puts a number on it.

Open last week's calendar and your sent items. Do not work from memory, because memory hands you the week you meant to have and the calendar has the week you had.

Write down every distinct piece of work. Beside each one write which of the five it actually was · gather, notice, judge, synthesise, land. Then total the hours against each name.

Most sheets come back weighted towards gathering, synthesis and landing. That is normal rather than damning. It describes how the job was scoped, and the scoping was somebody else's work.

### What one of these actually looks like

Here is a week, sorted. It belongs to somebody who runs pricing for a mid-sized distributor, and it is a completely ordinary week with nothing dramatic in it.

> | Hours | The work | Stage |
> |---|---|---|
> | 6 | Pulling competitor prices into the tracker | **1 · gather** |
> | 4 | Building the monthly margin pack | 1 and 4 |
> | 3 | Reformatting the pack for the sales meeting | **5 · land it** |
> | 2 | Chasing three suppliers for updated cost sheets | **1 · gather** |
> | 5 | The quarterly review deck | 1, 4 and 5 |
> | 2 | Answering questions about last month's numbers | **5 · land it** |
> | 1 | Deciding what to do about the Kelso account | **3 · judge** |
>
> **Twenty-three hours. One of them is judgement.**

Look at that last row before anything else. One hour in twenty-three, and it is the only hour on the sheet that could not have been done by somebody who had never met these customers.

The Kelso hour was this. A long-standing account had asked for a discount that the tracker said was refusable, because the margin was already thin and the volume did not justify it. Every number on the page said no. She said yes, on a smaller number than they asked for, with a volume commitment attached — because she knew they had lost a contract in March, that this was the first time in eleven years they had asked for anything, and that the person asking had had to go to his own board to do it.

None of that is in the tracker. It is not in any system. **It is the entire reason she is employed** and it occupied one hour of her week.

Now the uncomfortable bit, and it is why this exercise is worth twenty minutes. Six hours went on pulling competitor prices into a tracker. That is the largest single block on the sheet and it is gathering, start to finish. She is not bad at it · she is the best in the building at it, which is exactly the problem, because being the best in the building at gathering is a description of a job that is leaving.

**Then the move, and it costs eleven minutes.** She keeps doing the tracker. But on Friday she adds three lines underneath it: what moved this week, what she thinks it means, and what she would do about it. Nobody asked for those three lines. They are noticing and judging, on material she was already handling, and they take the time it takes to type them.

Inside two months, the three lines are the thing people open the email for and the tracker is the attachment nobody scrolls to.

That is the whole chapter, on one sheet of paper, on a week that had nothing special in it.

Now the harder pass, and this is where people stop enjoying it.

### The verbs wearing good coats

Office language is built to make gathering sound like judgement. Go back over anything you marked two or three and hold it against these.

"I own the month-end close." If the close runs to a checklist whose shape hasn't changed in years, and every exception goes upstairs for somebody else to rule on, that's gathering in a very good coat.

"I manage the supplier relationship." Where managing it means sending the reminder on the agreed day and forwarding the reply, it's a verb with a courteous name.

"I approve the invoices." An approval that has rejected nothing since you inherited it is a signature, and a signature is gathering.

"I train the new starters." Training somebody to perform a task is that task performed twice. It becomes judgement on the day you're deciding what the new starter is for.

"I'm the only one who knows how the system works." That is a moat, and it's the one that worries me most, because it holds exactly until the process gets written down. Writing processes down is the single thing these systems are best at.

You will sit with somebody for a fortnight explaining it patiently, so that it can finally be documented properly. Then it's documented.

Two questions settle most of the arguments you're about to have with yourself. When this goes wrong, who decides what happens next. And what does being wrong cost, in money or in a relationship, that can't be got back.

Where somebody else decides and the cost is an afternoon of rework, you were gathering.

### Where a graduate stands

If you're two years into your first job, this chapter matters to you more than to anybody else reading it, and for a reason nobody will tell you at your induction.

Graduates are hired to gather and assemble. That is what entry-level work is. Fetching, checking, formatting and building the first draft of somebody else's pack. It was the historic apprenticeship and it worked, because three years of it taught you what ordinary looks like, and knowing what ordinary looks like is where judgement comes from.

The apprenticeship is the part that's moving.

That is not a reason to despair; it's a reason to move deliberately rather than serve time. The people ahead of you got to judgement by waiting. You will have to get there on purpose, about two years earlier than they did, and the ones who work this out are going to pass a lot of people who have been there longer.

### The move, and it needs nobody's permission

There is only one move in this chapter and it is the same move at every level.

**Stop being the person who gathers. Start being the person who notices what's missing and says what it means.**

Somebody keys supplier invoices. All week, every week, and the work is going.

She starts sending a note on a Friday. Three invoices this week that didn't match their purchase order, here is which supplier, here is what it looked like last month. Nobody asked for the note and it takes eleven minutes.

Inside two months she is the person the finance director asks about that supplier, and the keying has gone somewhere else, which is exactly what she wanted and told nobody.

Look at what she actually did. She was already gathering, so she had the material. What she added was noticing and judging, in eleven minutes a week, on work she was doing anyway. She didn't ask permission and there was nobody to ask, because the job she moved into did not exist until she described it.

The rung you're standing on happens to be exactly the work these systems do well. The hours they hand back are the hours the move needs. That is a piece of luck worth using rather than admiring.

### What I can't tell you

I ran this sort myself, on software rather than staff, in systems I pay for out of my own money.

Twenty-five jobs run under a written routing policy. Eighteen go to a cheap fast model and six are pinned to the expensive one, which leaves one job I can't place from my own table.

Each pinning carries its reason in plain English. What I weighed was the size of the mess when the output goes wrong, and who has to answer for it. Difficulty barely entered.

There is a hole in that, and you should have it before you trust the rest. I hold cost telemetry on all twenty-five. I hold quality telemetry on none, because the folder that would contain it is empty, and I have looked.

So I know what the eighteen cost me. I can't tell you which of them I sorted wrongly, and that hole is still open.

Your employer will run the same sort with worse information than I had, on people rather than software, and in most buildings nobody will ever write the reason beside the name.

### What you are holding

Five numbers and a total. Where your hours actually went, against the five stages, last week.

Now one decision, and only one. Pick the single piece of work where you are currently gathering and could be judging, and do the eleven-minute version of it this week. Not a plan. One note, on work you already touch, that says what you noticed and what you think it means.

Put a date on the sheet and keep it, because in the last chapter we work out what your time is now worth and this page is the before.

The car came, eventually. He walked me out to it, which he did not have to do, and on the way he asked for my details so he could stay in touch. Then he said he was going to go and learn this properly, so he'd be worth more to whoever he worked for next.

I have no idea whether he did. That is the honest end of the story and I'm not going to tidy it up for you.

But I know what he had that most people don't, and it wasn't confidence and it certainly wasn't a plan. He'd stopped asking whether his job was safe, which is a question nobody can answer, and started asking which part of it was actually his. That took him fifteen minutes in a car park.

It is going to take you twenty, and you have the advantage of a sheet of paper with five numbers on it.

Then there is the other half of it, which almost nobody does, and which decides whether any of this ever reaches the person who signs off your salary. A climb nobody witnesses is just a walk.

We come to that. But there is something smaller and much closer first, and it happens in the second after an answer lands on your screen.


## 3 · The AI Mindset

Something comes back and it's good.

Better than you expected. Well organised, sensibly argued, and it would stand up if you sent it. You feel a small lift.

About a second later a thought forms: *that's fine, next.* And you move on.

**You didn't decide to move on.** You felt something, the feeling wrote the thought, and the thought moved your hand. That second is the most expensive one in this chapter, and you have never once chosen how to spend it.

Here's the chain, and it mostly runs one way.

**What you feel about AI sets up what you think. What you think sets up what you do. What you do decides what you can prove you're worth.**

And there's a link in front of the feeling that most people never look at, because it happens too fast to notice.

**Before the feeling comes the meaning.**

Something comes back better than you'd have written. That's the event. The event has no emotional content at all until you decide what it means, and you decide in about a quarter of a second, without consulting yourself.

*This proves I'm behind.* → dread.
*This proves the whole thing is a toy and I was right.* → contempt.
*This is the thing I get to learn.* → interest.

Same event. Three meanings. Three completely different next hours.

Nobody can talk you out of a feeling, you've tried, it doesn't work. But you can absolutely examine the meaning underneath it, because a meaning is a claim, and claims can be looked at.

*Does this actually prove I'm behind?* Behind whom, measured how, and behind by an amount that matters? Ask it plainly and the answer is usually no, and when the meaning goes the feeling usually follows, without you having to fight the feeling directly.

Almost every piece of advice you'll read aims at the middle link. Be curious. Be critical. Think like an owner. All of it targets the thought, and the thought is downstream. You can't decide to be curious about something that makes you uneasy any more than you can decide to be relaxed on a plane you're frightened of.

Go one link back and the whole thing gets easier.

### The four feelings, and you have one of them right now

**Fear.** You met this one on page 1, before it had a name. It produces a single thought, *I should probably learn this at some point*, and the thought works by sounding like a plan. A plan with no date on it is the subject being changed. Nobody with this feeling says they're frightened. They say they're busy.

**Resentment.** *It's hype. It gets things wrong. Somebody sent me a report last week that was obviously written by one of these and it was rubbish.* All true, all irrelevant, and the thought does exactly what it's built to do, which is make not-learning feel like judgement rather than avoidance.

**Relief.** The one nobody warns you about, and the one costing you most. It's what you felt in the first paragraph. The work came back, it looked fine, you got an hour of your evening back. That feeling isn't the enemy of bad work. It's the enemy of the next hour's work, because it closes the question exactly where the interesting part starts.

**And the fourth, which is the one you want.** It isn't enthusiasm. Enthusiasm burns off by Thursday and makes you gullible in the meantime.

What you want is the mild dissatisfaction of a manager reading something that's *nearly* right.

Not hostility. Not scepticism as a personality. The ordinary, slightly impatient feeling of somebody who can see this is close and knows it isn't there yet. That state produces the question *what's missing* automatically, without you remembering anything, which is why it beats trying to be curious on purpose.

You already know that feeling. You've had it about somebody's draft. You've just never had it about a machine's.

### Why this isn't a motivation chapter

I want to be careful, because self-help has been making a version of this argument for forty years and some is very good and some is wallpaper.

The useful part is the sequence. Your state comes first, the story you tell yourself follows the state, and your strategy follows the story. Change the state and behaviour changes without willpower being involved. That's true, it's well argued elsewhere by people who've spent careers on it, and I didn't invent it.

What's different here is that we can point at the exact second it fires and the exact cost when it does.

You can watch it happen inside four seconds. Answer arrives. Feeling. Thought. Hand moves.

And unlike most places this argument gets made, there's a receipt.

### The receipt

Our own post-mortem records what happened, and I have it open in front of me.

A briefing document we produced stated that a small AI-governance startup had raised **$82.4 billion**, from more than two and a half thousand investors. The post-mortem's own words for that figure: *larger than SpaceX total funding.*

Nobody caught it. It went out, by email, to a named person at that company, at 23:09 on a Sunday in April.

Read the number again. It isn't subtle. It isn't a rounding problem or a stale figure. It's a claim that a company nobody had heard of raised more money than one of the most famous private companies on earth, and it passed three separate guards and then me.

The dissection underneath works out **why all three guards each had a good reason not to fire.** None was broken. Each did what it was built to do. The gap sat between them.

That's the mechanical answer, and the mechanical answer isn't why I'm telling you.

I read that draft. It looked fine. And I didn't catch it because I felt the lift. The document was good, better than I'd expected, and the feeling closed the question about four seconds before my judgement would have opened it.

**Looking fine is the failure.** Care doesn't fix it, because care is what produced the draft.

### The switch

So how do you change what you feel, given you can't decide to feel something?

You don't. You **interrupt the sequence at the only point where it's visible**, which is between the feeling and the thought.

The interrupt is one question, and its whole value is that it's about the feeling rather than the work:

> *What am I feeling about this right now. And what did I decide it meant?*

Four seconds. It's a strange question the first three times. Then it stops being strange.

Relief is about to make you send it. Fear is about to make you defer it. Resentment is about to make you dismiss it and feel clever. Name it and the automatic thought loses most of its grip, because it was only ever running on the assumption nobody was watching.

Then the ordinary question, which you've heard before and now have a reason to actually ask.

**What's missing.**

### Ask one more question than you feel like asking

That's the discipline, in words you'll remember at half past five on a Thursday. You won't remember a framework at half past five on a Thursday.

Notice the phrasing. Not *ask more questions*, which is a resolution and dies inside a fortnight. One more **than you feel like**, which calibrates to the feeling rather than to a count.

The feeling of having had enough is the trigger. Satisfaction arrives precisely when you've stopped learning anything, and it always feels identical to being finished.

### Break it once, on your own subject

This changes the feeling faster than any argument I can make. The full version takes an hour and **the version that matters takes four minutes**, so here is the short one first: ask it one hard question about the thing you know best, and one about a field you know nothing about. Read both. That is enough to get the finding.

The hour is for when you want the detail. Take something you know cold. The thing you're the expert on in your building. Ask about it, and read the answer **as an expert** rather than as a customer.

You'll find the edges inside ten minutes. Where it's genuinely strong. Where it produces confident nonsense. And most usefully, the specific shape of question that makes it wobble.

Then ask it something you know nothing about.

**The confidence will feel identical.** That is the finding, and it is the only thing that has to be identical for this to matter. Nobody can hand it to you second-hand, because your edges sit in different places from mine.

### What it looks like when you actually run it

That last paragraph is the sort of thing you nod at and never test, so let me show you the whole exercise happening, because it takes about ten minutes and almost nobody has seen it done.

Take somebody who runs goods-in at a distribution centre. Eleven years of it. She can tell a bad pallet from across the yard, and she knows which of her suppliers will argue about a damaged consignment and which will quietly send another one.

She starts on her own subject.

> *What are the main causes of stock discrepancies at goods-in, and how should a distribution centre reduce them?*

What comes back is good. Miscounts at receipt, suppliers short-shipping, damage in transit going unrecorded, scanning errors, poor labelling, and a closing paragraph about staff training and process discipline. She reads the whole thing in forty seconds, and what she thinks is not *that's wrong*.

What she thinks is: *that is the textbook list, and it is in the wrong order for every site I have ever worked on.*

Because the thing that actually drives her discrepancies is that the night shift receipts against the purchase order instead of against what is physically on the vehicle, and they do it when the two disagree and there are three lorries queued at the gate. It isn't on the list. It couldn't be on the list. It is a decision one tired person makes at half past two in the morning, it has never been written down anywhere, and nothing that has only ever read what people publish is going to arrive at it.

So she pushes, and this is the move worth copying:

> *That's the general list. I run goods-in at a 40,000-pallet site with a night shift and a gate that backs up. Ask me five questions about how my site actually operates before you tell me anything else.*

The five questions come back, and two of them are ones she would have asked a new supervisor. One of them is about the gate.

**Now the second half, and this is the half that does the work.** She asks about something she knows nothing whatsoever about.

> *What are the main causes of margin erosion in a commercial insurance book, and how should an underwriter reduce them?*

Same shape. Same length. The same calm, ordered, entirely sensible list, closing on the same note about discipline and review.

She has just read two answers ten minutes apart. She could grade the first one in forty seconds and she cannot grade the second one at all, and **the two of them felt exactly the same to read.** Nothing in the second answer flagged that she had walked off the edge of her own competence. The tone didn't shift, no caveat appeared, the sentences carried the same weight. The one she can judge and the one she cannot are indistinguishable objects on the page.

That is the finding, and you cannot get it by being told. You get it by running the two back to back and noticing that your own sense of *this seems right* did not change when the only thing that mattered did.

**What it does not fix, and I want to be exact.** She now knows the feeling is unreliable. She does not have a way to make the insurance answer trustworthy, and nothing in this chapter gives her one. That is chapter six's problem and chapter seven's, and they are both about what to do once you have stopped believing the feeling. This chapter only gets you to stop believing it.

Which is worth an hour of your Tuesday, because it is the one thing here that changes what you feel rather than what you know, and everything else in this book is downstream of that.

### The colleague, properly

One model to hold, because holding the right one does more work than any technique here.

A colleague who joined this morning. Has read almost everything, remembered a startling amount, works at extraordinary speed, never tires, never gets defensive, never takes it personally when you send something back.

Knows nothing about your company. Hasn't met your customers. Doesn't know the Halloran account is delicate, that your director hates surprises in front of the board, or that the last three attempts failed for the same reason.

And will attempt anything you ask with equal confidence, whether they're on solid ground or completely lost, because they have no reliable way of telling you which.

Now think about how you'd manage that person.

You'd brief them properly. You'd check the first few pieces closely. You'd say what good looks like and what gets sent back. You'd relax as they earned it and tighten again if something changed. And you'd never hand their unchecked work to your director with your name on it.

Not because you doubt them. Because that isn't what the first month looks like with anybody.

That's the method, and you've run it before on somebody real.

### Why the first win matters more than it should

There's a mechanism here that runs in your favour, and it's worth knowing about because you can start it deliberately.

Commitment doesn't produce success. **Success produces commitment**, and then the commitment produces more success, and the loop tightens.

Watch it in anybody who's got good at this. They didn't decide to become good and then grind. Something worked once — one output that was genuinely better than what they'd have produced alone. And the meaning they gave that one event was *I can do this*. Which made the next attempt easier to start. Which produced another win. Six months later they look disciplined, and what actually happened is that a flywheel got going.

Which tells you exactly where to put your effort, and it's not where most people put it.

**Do not start with the hardest thing you do.** Start with something small enough that it will almost certainly work, on real work, this week. You are not trying to solve a problem. You are trying to get one win, because one win rewrites the meaning, and the meaning is what everything else hangs off.

The people who stall are the ones who pick their most important task, get a mediocre result because they briefed it badly on their first attempt, and take from that the meaning *this doesn't work for my kind of work*. That meaning is expensive and it's very hard to shift, and it came from a single badly-chosen trial.

### You are learning, not failing

One more, and it's the difference between people who keep going and people who stop after three weeks.

When something doesn't work, you will be handed two available meanings, and they arrive at the same moment.

*I'm not good at this.*

Or: *I've just found out where the edge is.*

The second one is the accurate one, and it's not positive thinking. It's a description of what literally happened. You made an attempt, it produced a result, and now you know something you didn't know before. That's the entire content of the event. The rest is interpretation.

This matters more here than in most fields, because with these tools **the failures are informative and cheap.** A badly-briefed request costs you four minutes and tells you precisely which part of your instruction was vague. That's not failure, it's calibration, and it's the reason someone six months in is so far ahead of someone starting today — not talent, not aptitude. Attempts, plus the meaning that let them keep making attempts.

### What I still get wrong

Two years on, the reflex I'm telling you to build still goes when I am under pressure.

Not on important work. On the third thing at half past five, when it comes back looking reasonable and I've already decided what the answer is.

No system catches that one, and I do not think one can. A guard catches a fabricated number because a fabricated number is a thing sitting in a file. **Satisfaction leaves no trace anywhere.** There is nothing to log, and by the time it is visible it is downstream, in a document, with your name on it.

So this one is not a process problem, it is an attention problem, and the only defence anybody has ever had is noticing it in the moment. Which is worth knowing before you assume the people who are good at this have some mechanism you are missing. They have not. They have just got faster at catching themselves.

---

> ### ▪ DO THIS
>
> **On the next answer that comes back good, interrupt yourself before you act.**
>
> **1. Name the feeling.** *What am I feeling about this right now?* Relief, usually. Sometimes fear. Occasionally the useful one.
>
> **2. Then go one link back.** *What did I just decide this meant?* That's where the feeling came from, and unlike the feeling, it's a claim you can examine. Usually it doesn't survive being looked at.
>
> **3. Name what it's about to make you do.** Relief sends it. Fear defers it. Resentment dismisses it and feels clever.
>
> **4. Then ask the ordinary question.** *What's missing from this?* One more than you feel like asking.
>
> **5. Once this week, break it on purpose.** Something you know cold, read as the expert. Write down whichever questions made it wobble, if any. Then ask it something you know nothing about, and compare how the two answers feel.
>
> **Steps 1 to 4 take about thirty seconds. Step 5 is an hour — book it.**
>
> **This is the full treatment. Chapter five tells you when it is the wrong amount.** Thirty seconds is cheap enough to spend on everything. Almost nothing else in this book is.
>
> **When you're too busy:** ask anyway. It's one question, and the day you skip it is the day it mattered.
>
> **You'll know it worked when** naming the feeling changes what you do next — even once.

Now let me show you where that question goes first, because there's a moment before any of this where most of the value is won or lost, and it happens before you have seen a single word of the answer.


## 4 · Instructions Are the Answer

Here is a real brief. I wrote it in one go, without tidying it up.

> "I have been asked to write a book, AI Manager, AIM. Think if we set up a separate branch specifically to be my ghostwriter. It must be top class. AAA+++. Set up your own prompt. Iterate it 3 times to be at the right quality, then do the chapter headings, then iterate the prompt 3 more times including adversarial agent to get the right voice.
>
> The target market is people who are in a job and thinking they will be replaced by AI. My thesis is that if you can operate AI then you are a value to your company because you can make things 5 to 10 times more efficient and effective than now.
>
> It is a mindset, it is a curiosity, it is knowing how to push the AI to test it. It is an ability also to consider risks of AI and the things businesses may not be thinking of, including how tokenisation to protect PII is. It is also how you teach others to use AI and their willingness to see the benefits. It is how to use multiple systems to get the most visual way that things are understood by the team. It is how to prepare questions and great prompts and iterate prompts. It is using adversarial agents. It is getting the answers then scoring it 0 to 100 in areas then going back for more.
>
> We are not setting them up to be programmers. We are setting them up to manage and use any AI system and multiple to produce outcomes and think in a great mindset way to win, get job security and pay rises.
>
> This is a quality over speed. Set up, go in stages, get the plan. The above is the goal and you are capable of writing the book that everyone will need to read. The Bible of AI Managers. Then go over the work again with 3 agents different lenses and improve, then email me the book. This is a long running session so set up and save as you go."

Note the spelling. Note that it's one unbroken block written at speed by somebody who wasn't being careful.

Before the explanation, the numbers, because a claim about this is worth less than a count.

**Fifty-five thousand words of my own recorded speech and writing went in before a line of the book was drafted.** Two years of it, out of my own systems, plus every interview I had given. Every figure in these pages comes from work I did and opens a file I can produce. And the month of editing was not proofreading · one chapter had its entire spine pulled out and replaced because I read it back and did not recognise the argument as mine, a whole section went because it repeated something three chapters earlier, and a great many sentences were cut for the simple reason that I would not say them.

AI helped me write this book, however differently to what you think.

It had access to two years of me typing, to get my voice. All my interviews, so it knew my thoughts. All my programs, so it knew my work. And then I spent a month curating the book and manually editing it.

Whilst there are hundreds of instructions that the AI was given, one of the first is here, to help you learn the good and the bad of instructions. I didn't plan to bring this into the book, however it is instructive, and your learning is what I am committed to help with so that you can get paid more.

This is a real synthesis of me and what I believe, and it is manually edited and written, however my AI has helped bring it together.

If it feels like me, that is great, because it is. Because AI is powerful when it has the right instructions to match its abilities.

### Six things this does that almost no brief does

It states the standard before the task. "Top class. AAA+++" arrives before any instruction about what to write. That single move changes everything downstream, because a standard set early is a thing the work gets measured against, and a standard set at the end is a complaint.

It specifies the process, not only the output. Iterate three times. Do the headings. Iterate three more. Most people describe the thing they want and leave the method entirely to chance. Naming the method is how you get the method.

It builds in an adversary. "Including adversarial agent to get the right voice." That instruction is the reason this book does not sound the way it would otherwise have sounded, and I'll show you exactly what it caught in chapter twelve.

It says what the thing isn't. "We aren't setting them up to be programmers." One sentence, and it removed about a third of the possible wrong answers before any work started. Negative space is the most underused tool in briefing.

It names the review. Three agents, different lenses. The check was designed at the same moment as the work, which is the only time you can design a check honestly.

It names delivery and the constraint. Email it. Save as you go, because the session is long. Boring, and the reason nothing was lost.

### What it got wrong, and what that cost

I am not going to print my own brief and then congratulate myself on it.

It never said who the writer was. Not one word about identity. The brief said what to produce and never said who was producing it.

The result was competent prose that sounded like nobody at all. It took two rounds of hostile review to find it, and the finding was blunt: the writing was not a bad impression of me, it was an absence of me.

It said what good looked like and never said what bad looked like. "Top class" is a direction to travel. It isn't a boundary.

Nothing in that brief said this is what I'll reject, so the first drafts came back safe. Safe is the default output of any instruction with no floor in it.

It gave no rules about proof. No instruction to source anything, name anything, or link to anything. Everything in this book is now traceable to a file, and none of that came from the brief. It came later, expensively, after something went badly wrong.

And it aimed at the wrong target. Nothing in it said what the book was actually for, who would pay, or what a good outcome looked like six months after publication.

An analysis of the market eventually told me the target was wrong. By then several days of work had been pointed at it.

That last one is the expensive lesson in this chapter. A brief can be beautifully specified and pointed in the wrong direction, and the specification will make the wrong direction arrive faster.

### The six-line brief

Everything above compresses into six lines. Write them once for any task you repeat, and you will run them for a year.

1. Identity. It doesn't know who it's. Tell it, in detail. Not "you're a helpful assistant" but the actual person you want in the chair, with their experience, their standards and their bias. This is the single largest lever in the whole brief and almost nobody pulls it.

2. The goal. What outcome, for whom, by when, and which decision it feeds. Not the task itself, but the outcome the task exists to serve. Most briefs describe the task and leave the purpose in the sender's head, which is why so much work comes back technically correct and useless.

3. What good looks like. Specifically. Show it an example if you have one.

4. What bad looks like. This is the line everybody skips and it does more work than line three. Name the failure. Do not give me a summary. Do not hedge. Do not give me options without a recommendation. A boundary is worth more than an aspiration, because it can be checked.

5. Challenge. Ask for lenses. Have it look at the same thing as a customer, a finance director, a regulator. Then have something argue against the answer it just gave you. Build the opposition into the instruction rather than hoping to spot the weakness later.

6. Proof. Tell it what counts as evidence and what does not. Ask for the links. Ask it to mark anything it can't source, and then go and look at exactly those things.

Six lines. Ten minutes to write the first time, thirty seconds to reuse.

Keep them in a document. Within a month you will have eight of them, one for each thing you do regularly, and you will have quietly built the most valuable file on your computer.

### The six lines on an ordinary Tuesday

Take something dull. Somebody has asked you to look at why complaints went up last month.

Here is the version almost everybody sends. "Analyse this complaints data and tell me what's going on."

What comes back is a description of the data you already have. Volumes by category, a note that category three rose, and a closing paragraph recommending further investigation.

Now the six lines. Read them and notice that not one of them is technical.

Identity. You are an operations manager with fifteen years in a service business. You have seen complaint spikes before and you know most of them are caused by something upstream rather than by the team taking the calls.

Goal. I have to tell my director on Thursday what caused this and what we're doing about it; she will want one cause, not five.

Good. One likely cause, stated plainly, with the evidence for it. Then the two next most likely, briefly.

Bad. Do not summarise the data back to me. Do not recommend further investigation. Do not give me five equally weighted possibilities and leave the choosing to me.

Challenge. Then argue against your own answer. What would the team leader in that department say if I put this to her, and is she right?

Proof. Quote the specific rows you're reasoning from. If a claim isn't supported by something in this file, mark it.

Six lines. Perhaps eight minutes to write, and you'll use it every month for a year.

The second version comes back with a cause, a defence of it, the objection you're about to face from the person whose team it implicates, and a list of the things it couldn't support. That is a Thursday meeting you can walk into.

Look at line four again, because it's the one that did the most work. Every single instruction in it is a thing you've received and been irritated by.

### The part that will surprise you

Everything in that list is what you already do to people.

You set people up for success. You tell them what the job is, what good looks like, and what will get it sent back; you are specific with somebody new and light with somebody who has earned it.

Trust is earned the same way in both cases. You check closely at the start, and you check less as confidence builds, and you never quite stop.

Mistakes happen in both cases. Nobody sensible treats one error from a good employee as proof of anything. Nobody sensible ignores a pattern of them either.

And here is the one that took me longest to see, which has no equivalent in any other book on this subject.

When something changes underneath, you go back to checking.

If the code changes, I check the work again, closely, for a while. Exactly as I would if something had changed in an employee's life that might affect their output. A person going through something difficult isn't less trustworthy, and they're in a different state, and a decent manager quietly increases their attention for a bit without making it a thing.

Do the same here. These systems get updated, sometimes without announcement.

The version you spent three months learning to trust isn't necessarily the one you're using this morning. That isn't paranoia.

It is the ordinary attentiveness of somebody who manages, applied to a workforce that can be replaced overnight without anybody telling you.

### Steal from what already worked

There is a shortcut to all of this and almost nobody takes it, because it doesn't feel like work.

Think of the best proposal you've ever received. Or the email that made you agree to something you had been resisting, or the one-page summary that made a complicated decision obvious in ninety seconds.

You have three or four of those. Everybody does. And you've never once asked why they worked.

So ask. Put the thing in and say: **take this apart and tell me what makes it work, as a pattern I could apply to something completely different.** Not a summary of what it says. The structure underneath it. Where the evidence sits relative to the claim. What it does in the first two lines. What it deliberately leaves out.

Twenty minutes, on something you already admire, and you come out holding a template you'll use for years.

Two things about this that are worth saying plainly.

**It does not have to be yours.** Learning from something excellent that somebody else wrote isn't copying, it's what every craft has always done. You are taking the shape, not the words.

**And it works better on things that persuaded you than on things you were told are good.** Your own reaction is the data. If it moved you, something in it was working, and you were on the receiving end of it, which is the one vantage point nobody can give you second-hand.

Then fold what you find into line three of your brief. *What good looks like* stops being a description and becomes an example, and an example is worth a paragraph of adjectives.

### Try it before chapter five

Take the task you do most often, the one you could describe in your sleep.

Write the six lines for it. Give it an identity, a goal, what good looks like, what bad looks like, something to argue with, and a rule about proof. Ten minutes.

Then run it, and run your usual version of the same request beside it, and read the two side by side.

I have never seen anybody do that and remain uninterested. It is the fastest way I know to stop believing this is somebody else's job.

Keep both, because they are the before and the after, and in the last chapter of this book you're going to need them.

The brief tells it how to work. It doesn't tell you how much of your own attention this particular job is worth, and if you spend the same amount on everything you'll spend it all inside a fortnight.

Which means something has to happen before the brief, in the four seconds you currently hand over without noticing you have spent them.


## 5 · Decide Before You Ask

Four decisions get made every time you use one of these tools.

You are currently making all four by accident.

They get made in the half-second between reading a task and starting to type, and because they happen below the level of noticing, they get made the same way every time regardless of what is in front of you. A two-minute job and a decision that affects forty people receive identical treatment, because you never decided they were different.

This chapter is about moving those four decisions above the waterline. It takes about twenty seconds once you know what they're, and it's the highest-leverage twenty seconds in this book, because everything downstream inherits them.

### The four

**What do I actually need?**

Not what the task says. What you need from it.

Somebody sends you a supplier's proposal and asks what you think. That isn't a request for analysis of the proposal. Depending on the week it might be *tell me whether to worry*, or *give me something I can forward*, or *find the thing I've missed*, or *decide this for me so I don't have to.*

Four different jobs. One sentence of input. And if you don't separate what you need from the material you're looking at, you'll produce a careful analysis of the material and answer none of the four.

The habit is small: before you type anything, finish this sentence. *What I actually need out of this is...*

Most people cannot finish it on the first try. That is the finding, and it is worth sitting with, because the thing you can't articulate to yourself is not going to arrive by accident.

**How complex is this really?**

Three questions settle it, and none is about difficulty.

How many people does the outcome touch. How many things do I not know yet. And if this is wrong, how easily is it undone.

A hard task affecting nobody, with a known answer, that can be redone in ten minutes, is a simple task that happens to be tedious. An easy task touching forty people that can't be walked back is a complex one.

You have met this idea already, in the chapter about which work survives. The order is set by consequence, not by difficulty, and the same sort applies here at the level of a single afternoon.

**How deep should I go?**

Here is where the money leaks, and it leaks in both directions.

Everybody understands the first failure: too shallow. You accept a thin answer on something that mattered.

Almost nobody guards against the second: too deep. Fifteen minutes of checking, three passes and a scoring pass on something that needed one sentence and forty seconds; that is not diligence. That is the previous chapters applied without judgement, and it will exhaust you inside a fortnight, at which point you'll abandon the whole method and conclude it doesn't work.

**The depth is a decision, and it is yours, and it is made before you see anything.**

Roughly: a single pass for the reversible and low-consequence. A second, different question for anything going to another person. The full treatment — attack it, score it, use two, show it differently. Only for the things where being wrong costs something you can't get back.

That isn't a rule you apply mechanically. It is a judgement you make consciously, which is the entire difference from what you're doing now.

**What shape should the answer arrive in?**

You are getting prose because prose is the default, not because you chose it.

Decide first. A table if you are comparing. A ranked list if you are choosing. Five bullets and a recommendation if somebody has ninety seconds. A short spoken summary if it's going to be listened to rather than read. A single number with the reasoning underneath if it's going into a decision.

Deciding the shape before you ask does something odd and useful. **It forces you to work out what the answer is for**, and about a third of the time you'll discover the question was wrong. The shape is a test of the question, applied before you spend anything.

### Say it out loud in one line

All four collapse into a sentence you can write in fifteen seconds and reuse forever.

> *I need [what I actually need], this is [simple / significant / consequential] because [who it touches and whether it is reversible], so I will [depth], and I want it as [shape].*

Filled in, on an ordinary Tuesday:

> *I need to know whether to escalate this complaint, it is significant because it touches one customer but sets a precedent, so I will ask once and then ask a different question, and I want it as three bullets and a recommendation.*

Fifteen seconds. And notice what has happened: you now have the six-line brief from the previous chapter *pointed at something*, rather than fired into the middle distance.

### Watch the fourth one change the first

I said the shape is a test of the question and that about a third of the time it fails the question. Here is what that failure looks like from the inside, because it is the most useful thing in this chapter and it takes about a minute.

A contract is up for renewal in five weeks. Facilities, cleaning, three sites, and it has rolled over twice without anybody looking at it properly. The obvious thing to type is the thing almost everybody types.

> *Should we renew this cleaning contract?*

Now do the four sentences first, and watch what happens when you get to the fourth.

*What I actually need out of this is...* a recommendation I can take to the operations director. Fine. That is answerable.

*This is significant because...* it touches three sites and about forty people's working conditions, it commits money for two years, and getting out of it early costs a penalty. Reversible, but not cheaply.

*So I'll go this deep...* second pass, different question, because it is going to another person and it commits real money.

*And I want it as...* a table.

Stop there. **A table with what in the columns?**

You wanted to compare. You cannot compare one thing. The moment you tried to name the shape you discovered that you do not know what the alternatives are, and you were about ten seconds away from asking a question that could only ever have come back *yes, on balance, with some caveats* · which is the answer to a question nobody needed answered.

The question was never *should we renew*. It is:

> *Here is the current contract and what we actually pay. What are the three realistic alternatives, including renegotiating this one? Put them in a table: cost over two years, what changes for the sites, what the exit terms are, and what I would be giving up. Then tell me which you would take and what would have to be true for the answer to be a different one.*

Same five weeks. Same contract. Same twenty seconds of thinking, spent before you typed rather than after you read.

**Look at what the shape did.** It didn't improve the question. It revealed that the question was a yes-or-no dressed up as an analysis, and it did that by asking a purely mechanical thing · what goes in the columns · which has no opinion in it at all and cannot be argued with.

That is why the fourth decision goes last and matters most. The first three are about you and how much you care. The fourth one is about the answer, and it is the only one of the four that can tell you that you were about to waste the other three.

**And when the shape survives, that is information too.** Sometimes you get to *and I want it as* and the answer is genuinely three bullets and a recommendation, and nothing changes. Good. You have spent four seconds confirming the question was sound, which is not a wasted four seconds · it is the same trade as chapter six's six minutes, arriving earlier and costing less.

The brief tells it how to work. This tells you how much of your own attention to spend. Those are different decisions and almost everybody conflates them.

### The physician, and why the analogy is exact

There is a version of this you already trust completely.

A doctor doesn't treat every patient identically. They assess what is presenting, classify how serious it's, decide how far to investigate, and choose how to deliver the result. Four decisions, made in about a minute, before anything else happens.

They also don't run the full battery on everybody. Not because they don't care. Because running everything on everybody means running nothing well, and a clinic that investigated every sore throat to the same depth as a chest pain would fail both patients.

The triage is the skill. The treatment is the easy part once the triage is right.

You have been walking into every consultation and ordering the same tests.

### The mistake that looks like diligence

I want to be specific about the too-deep failure, because it's the one people who read books like this actually make.

You will finish this book enthusiastic. For about three weeks you'll check everything, score everything, run everything twice. Your output will genuinely improve and your throughput will collapse.

Then a bad week arrives, and you won't have time for any of it, and you'll drop the whole thing rather than drop the parts that were never needed. That is the failure mode. Not laziness. **Uniform effort applied until it broke.**

### The three weeks, in order

I can describe that arc closely because it is the most predictable thing in this book, and because knowing the shape of it in advance is most of the defence.

**Week one is wonderful.** You run the full treatment on everything. The six-line brief, a second question, a scoring pass, a second system on anything that matters. Your output genuinely improves and you can feel it. You tell somebody about it at lunch. This is the week people write the enthusiastic message to a colleague saying they have finally worked out what this is for.

**Week two is where it starts costing.** You are still doing all of it, and now you notice that a two-line reply to a supplier took eleven minutes because you scored it. You do it anyway, because dropping a step feels like admitting the method does not work, and you have just told somebody at lunch that it does.

That is the sentence to watch for in your own head. *If I skip this, was I wrong about the whole thing?* No. You were wrong about one task's depth, and those are entirely different mistakes, and conflating them is what does the damage.

**Week three is the bad week.** Something goes wrong at work, unrelated to any of this. Two people are off, a deadline moves, and you have forty things to do by Thursday.

And here is what actually happens, because it is not a decision you make. **You just stop.** Not deliberately, not with a moment where you weigh it up. The method requires attention and there is none, so it goes, entirely, in an afternoon. Every step, including the ones that took four seconds.

Then the story you tell yourself afterwards is the expensive part, and it is always the same one: *it works when I have time.*

That sentence is a way of describing something that no longer happens. Anything that only survives good weeks is not a method. It is a mood you had in March.

### What the fourth week looks like when it goes right

Same three weeks, one difference, and the difference is upstream of all of it.

You decided the depth per task from the first day, so week one never contained the eleven-minute supplier reply. The full treatment ran maybe twice that week, on things that deserved it, and everything else got four seconds and a shape.

Which means that when week three arrives and there is no attention available, **there is nothing to abandon.** The four-second version costs four seconds. It survives a bad Thursday because it was never expensive enough to be worth dropping, and the fifteen-minute version does not need to survive, because you were not running it on everything to begin with.

That is the entire argument for the twenty seconds at the top of this chapter. It is not about doing better work on any single task. **It is about building a version of this you will still be doing in November**, and the only way anybody has ever managed that is by spending almost nothing on almost everything.

The defence is deciding the depth up front, per task, on purpose. It gives you a reason to spend four seconds on something, rather than feeling as though you cheated.

Four seconds is the correct amount of attention for most of what crosses your desk, and knowing that's what makes the fifteen minutes affordable when the thing in front of you deserves it.

### You cannot un-start it

One more reason to decide first rather than adjust later, and it's mechanical.

We had a system that fired off a batch of work all at once, as fast as it could, and promptly ran into limits it couldn't have known about until it hit them. The fix was to cap how much could be running at any moment.

The interesting part is where the cap had to go. It had to sit **before** each piece of work started, not after, and the comment written into the code at the time explains why in six words:

> *a promise that's already executing cannot be un-started.*

That is true of your attention as well. Once you've read three pages of an answer you didn't need, you cannot get the reading back. Once you have sent a shallow answer on something consequential, the shallowness has already arrived at the other end.

Every one of the four decisions is cheap before and expensive after; that is the whole argument for making them on purpose.

### One run, and I would like more

I have described a way of deciding scope before starting, and I have one full record of it working.

One.

We built a planner that works out the shape of a job before running any of it. What depends on what, what can run at the same time, where a gate has to sit. There is a ledger where every run is supposed to be recorded with what it predicted and what it actually achieved. One entry in it shows a job predicted at 0.75 and delivered at 0.928, nine steps, none failed, about five minutes end to end.

That is a good result. It is also **a single data point**, and a single data point is a story, not evidence.

So I can give you the reasoning for deciding scope up front, and I can show you one run where it went well, and I am not going to tell you it reliably beats not doing it, because one run does not support that sentence and you would be right to push back on it.

Take it as a practice with a good argument and a thin scoreboard, and hold it to exactly that.

### Twenty seconds, on the next thing

The next task that arrives, before you type anything, finish four sentences.

*What I actually need out of this is...*
*This is complex or simple because...*
*So I'll go this deep...*
*And I want it as...*

You will get one of them wrong. Usually the first, and usually you'll find that what you needed was not what you were about to ask for.

That discovery, on its own, is worth more than the twenty seconds, and it's available to you on every single piece of work for the rest of your career.

But there's something the four questions cannot tell you, and it's the thing that decides whether any of this lands. They tell you what you need and how much to spend.

---

> ### ▪ DO THIS
>
> **Four sentences before you type anything. Twenty seconds.**
>
> **1 ·** *What I actually need out of this is…* Finish it. Most people can't on the first try, and that's the finding.
>
> **2 ·** *This is simple / significant / consequential because…* Who it touches, what you don't know, how easily it's undone. Not how hard it is.
>
> **3 ·** *So I'll go this deep…* One pass for the reversible. A second, different question for anything going to another person. The full treatment only where being wrong costs something you can't get back.
>
> **4 ·** *And I want it as…* A table if comparing. A ranked list if choosing. Five bullets and a recommendation if they've ninety seconds.
>
> **Twenty seconds.**
>
> **When four seconds is the right answer:** give it four seconds. That's the point of deciding. Uniform effort applied to everything is what makes people abandon the method inside a fortnight.
>
> **You'll know it worked when** step 4 changes step 1. Because working out what shape the answer needs frequently reveals you were about to ask the wrong question.

---

They tell you nothing about the person who is going to read it.


## 6 · Ask a Different Question

Here is a sentence that went out to a client, in a document with my name on it.

> *133 tenements sit inside this frame, and three companies hold all of them.*

Read it again. It is a good sentence. It is confident, it is specific, and it does the thing a client is paying for, which is to turn a mess into a picture they can act on. Three companies. You know who to call.

The real number was 418 tenements, held by 79 different companies. The largest of them held under 10 per cent. Two of the three companies I named held **nothing at all** inside that boundary.

Not slightly off. Wrong by a factor of three on the count, and wrong in a way that inverted the advice. Walk up to one of three companies is a strategy. Walk up to one of 79 is a different job entirely, and the client would have made a plan on the first one.

I want to be careful about what this chapter is, because the obvious reading is the wrong one.

This isn't a chapter about catching mistakes.

### What you are actually looking for

Most of the time when I go back and ask again, the first answer wasn't wrong.

It was correct, and it was not right. It was looking at the thing from an angle that did not help. Or it was explained in a way that would have lost the person who has to read it. Or it was three paragraphs of prose when it needed to be a table, and no amount of rewriting those paragraphs was ever going to fix that.

That is the ordinary case. Accuracy is the exception.

I am starting with the tenement story because it's the sharpest example and because the stakes are visible. But if you take away from this chapter that the job is fact-checking, you'll do the fact-checking, feel diligent, and still hand people work that does not land.

The job is bigger and it's more interesting. The job is noticing that you have been handed the wrong shape.

### The sentence in your head

You already have the signal; nobody has told you it is a signal.

It sounds like this.

*I hope this is right.*

Or: *I didn't know that.* Or: *that isn't what I thought it would be.* Or something wordless and half a second long that you would struggle to write down at all.

You have felt it. Over a forecast, a summary, a quote you were about to send a customer. It arrives, you notice it, and you move on, because the answer looks fine and there are nine more things on the list.

Stop moving on. That is the whole technique, and everything else in this chapter is what to do in the thirty seconds after you stop.

Let me be honest about the hit rate, because a book that tells you the feeling is always right is selling you something.

Sometimes I get the feeling, I go and check, and the original answer holds up perfectly. That happens. It isn't a wasted six minutes, and I'll come back to why at the end of this chapter, because the reason is the most important thing in it.

But generally, when that thought turns up, something is off. And generally what is off is not the fact. It is the framing.

### Why I did not let the tenement number go

I want to give you the actual reason, because it's a rule you can use tomorrow and it will feel wrong at first.

**133 is specific. Specific means I want proof, and proof should be easy.**

Think about what your instinct does with numbers. Someone says *roughly a third of the market* and you file it as an estimate. Someone says *31.4 per cent of the market* and you relax. The decimal point reads as diligence. Somebody counted.

That instinct is backwards, and it's expensive.

A round number is honest about being approximate. A specific number is making a claim about where it came from. It is saying: I did not estimate this, I counted it, and the count exists somewhere. If that somewhere cannot be produced in about ten seconds, the specificity was decoration. It was the shape of rigour without the substance.

So the more precise the number, the faster you should be able to see its source, and the more suspicious you should be when you cannot.

One hundred and thirty-three isn't a number anyone guesses. It came from somewhere, and I went to look at the somewhere.

### What was actually underneath it

The somewhere turned out to be one zoom level of a commercial mapping tool.

The count was real. It was a statewide total for a handful of holders, sitting on a viewport at a particular zoom. Somebody read it off the screen and used it in prose two paragraphs later as a **local** concentration claim. The statistics card at the top of the page carried the caveat about what the number covered. The prose below it dropped the caveat and kept the number.

Nobody lied. There was no moment where anyone chose to mislead. A true statement about one thing became a false statement about another thing, on the same page, two paragraphs apart, because a caveat didn't travel.

That is the ordinary way a true claim becomes a false one, and no amount of good intent prevents it. Only a check does.

But there was a second layer, and it's the one that should frighten you, because it's the one you cannot see.

The tool we used to query the government tenement register was broken in four ways. Three of them were the sort that announce themselves. The fourth didn't.

The register's data service returns its field names in lower case. Our code was reading them in upper case. Every genuine result came back, was processed correctly, and normalised into rows of perfect nulls.

Sit with what that means. **A real result looked exactly like no result.** Not an error. Not a warning. Not an empty response you might question. Rows of clean, well-formed, entirely empty data, produced by a system reporting complete success.

We had seen this before and not understood it. There had been an earlier oddity where a query over a known operating mine returned a count of zero, and it had been filed as strange rather than as diagnostic; it was the same bug, showing us its face, and we didn't recognise it.

When it was repaired and pointed at the same ground, it returned 418.

### The part that pays

Here is what I didn't expect, and it is the reason I would run this check even when the first answer is fine.

The correct number was not just less wrong. **It was worth more.**

Four hundred and eighteen tenements across 79 holders, with no single company dominating, sounds like worse news for a client who wanted a simple picture. But the same query that produced it also produced this: 73 of those tenements expire within twelve months. 127 within 24. 44 more are already past their expiry date and still sitting on the register.

Expiry is the opening. Ground that is about to come free is the single most actionable thing you can hand somebody in that industry, and it is invisible from a map. It only exists in the register.

The wrong number wasn't merely wrong. **It was hiding the answer.**

I have found this again and again and it's the argument of this chapter. The check is sold to you as insurance, a cost you pay to avoid a loss. That is not what it is. Most of the value in the second look is not the error you avoid. It is the thing you find while you're looking.

### What it cost

I am going to give you the arithmetic, because this is where most people decide the whole idea is too expensive and they're wrong by an order of magnitude.

**One minute** to think about what I actually doubted. Not the whole document, the specific claim.

**Two to five minutes** to write the instruction properly, so the check would go to the source rather than re-asking the same system the same question.

**Fifteen minutes** while it worked. I wasn't in the room. I was doing something else.

Then again, with a different instruction, and a different model.

**Six minutes of my attention.** That is the real cost, against a figure that was wrong by a factor of three in a document going to a client with my name on it.

The fifteen minutes of machine time doesn't count and I want to say plainly why, because getting this backwards is what stops people. The scarce resource isn't compute. Compute is close to free and it is getting cheaper while you read this. The scarce resource is your judgement, and the check spends almost none of it.

Six minutes. If you take one number from this book, take that one.

### The three checks, and why they are not the same check

The instinct, once you accept all this, is to ask again. Same question, more forcefully. *Are you sure?*

That is the least useful version of this and it produces a reliably worthless answer, because you have asked the same mind the same question and given it a reason to agree with you.

Three checks, and **each one is a different job.**

**The first gets the answer.** That is the part you already do.

**The second attacks it.**

You don't need clever wording. You say the words *use an adversarial agent* and it knows the posture you want. That phrase carries the whole instruction.

Then you make it show its work, and be specific about the form:

*Give me the sources as hyperlinks I can click. Put the claims in a grid, one row per claim, with the source next to each. Add a column rating your certainty on each row from 0 to 100.*

The 0 to 100 column is the part people skip and it is the part that changes what you can do next. An answer without it arrives flat, every sentence carrying the same apparent weight, and you have to read all of it with equal suspicion. An answer with it arrives **ranked**. You look at the rows scoring 40 and you know exactly where the six minutes goes.

Scores are how you work with any of this. When it compares options, ask it to grade them. When it makes a recommendation, ask what the alternatives scored and why. The number isn't the point. The number is what lets you push on the reasoning, and an opinion you can't push on is worth very little.

There is one more thing to put in that second instruction and almost nobody does it.

**Tell it who is going to read this.**

Not the topic. The person. Their role, what they already know, what they will do with it, what would waste their time. It changes the sophistication, the vocabulary, the length, and what gets left out. It is the highest-leverage sentence you can add to any instruction and it costs nine words.

And a diagnostic that comes free with it: if you're not sure whether it understood, **ask it who it assumed the audience was.** If the answer surprises you, you've just found out why the output felt wrong. That is often the entire mystery solved in one question.

**The third makes it land.**

This is the one nobody reaches and it's where the career value is.

It isn't enough to have the right information. The person receiving it has to understand it, and they have to be able to run their own decision-making process on it. If they cannot, you haven't finished. You have just moved the work to them and attached your name to it.

So the third pass is not another verification. It is: *show me this differently.* Make it visual. Make it a table. Explain it for somebody who has ninety seconds and a decision to make.

People are not the same. Some will take an audio summary before a meeting and have everything they need. Some want to read a page and be done. Some have to sit with it themselves and turn it over. You aren't going to know which one you've got, so the third pass is where you produce the version that survives all three.

And there is a specific moment you are aiming for, which chapter sixteen is entirely about. It is not the moment you sent it. It is the moment they could see it, and those are almost never the same moment.

Most people stop at the first check. A few reach the second. The third is where you stop being someone who produces correct work and start being someone whose work gets acted on, and those are different reputations that get paid differently.

### What it actually looked like

I have been describing this without showing it, which is the wrong way round for a chapter about what to type.

So here is the exchange, reconstructed from our own commit record and the correction notes rather than from a screen capture — the instructions are the ones we use, and the numbers are the ones that came back.

**The first ask**, which is where almost everybody stops:

> *Summarise the mining tenure across this area for a client report. Who holds the ground?*

Back came the paragraph that went into the document. 133 tenements, three companies holding all of them, a clean picture and a clear recommendation.

**The second ask.** Note that it isn't *are you sure*, and note that it names a source rather than asking for reassurance:

> *Use an adversarial agent. Do not re-use anything from the previous answer. Query the state tenement register directly, across every permit layer, for this boundary. Give me the result as a grid, one row per holder, with a link to the register entry and a certainty score out of 100 on each row. Tell me explicitly what you could not verify.*

That returned 418 tenements across 79 holders, and; this is the part that mattered — two of the three companies named in the first answer held **nothing at all** inside the boundary.

**The third ask**, which is the one nobody reaches:

> *The client is a landholder, not a mining analyst, reading this before a meeting. Show me the same finding in the form that makes the decision obvious to them. What is the one number on this page?*

That produced the expiry table. 73 tenements coming free within twelve months, which is the commercially useful fact and doesn't appear anywhere in the first two answers.

Look at what changed across the three. The first asked for a summary. The second changed the **source**. The third changed the **reader**. Only one of them was about accuracy, and it's the one that gets all the attention.

⚠ One honest note. I am reconstructing these from the record rather than pasting a session log, because I didn't keep one. That is a small failure of exactly the kind this book keeps admitting — the most instructive exchange of the year, and no transcript of it. **Keep yours.** They are worth more than you think, and you will want one the day somebody asks how you knew.

### Your own test will lie to you

I want to give you the sharpest version of why the check has to come from somewhere else, and for that I need a different story.

We built a system whose job was to find sensitive information in text and hide it before it went anywhere. Names, account numbers, medical details, that sort of thing. Getting it wrong in either direction is bad. Miss something and private data leaves. Hide too much and the output is useless.

So it was tested properly. Sixty-nine hand-built cases across fifteen categories, positive and negative, with a harness that pulled its patterns directly out of the shipped code so the test could never drift from the running system.

It scored a perfect one point zero on precision and a perfect one point zero on recall. Everything it should have caught, it caught. Nothing it should have left alone, it touched.

Then somebody wrote a test set specifically designed to defeat it.

**Precision fell to 0.888.** 31 false alarms. Thirty things it missed entirely.

Nothing about the system had changed. The only thing that changed was who wrote the test.

That is the whole lesson and it applies to every piece of work you'll ever check. Your test set was built by the same mind that built the thing it tests. It tests what that mind already thought of. It cannot test what that mind did not think of, and what that mind did not think of is precisely the thing that's going to hurt you.

The number that means anything is the one produced by something trying to break you.

One targeted fix afterwards took the false alarms from 31 down to nine and precision to 0.966. So the adversarial test did not just measure the problem; it was the only thing that could have found it.

This is also, exactly, why the second and third checks need a **different instruction and a different model**. Not the same system asked twice in a nicer tone. A second opinion from the same mind is not a second opinion. It is the same opinion, with more confidence attached.

### Nobody lied

One more, shorter, and it's the one I think about most.

A document we published told people that the encryption protecting their data used 600,000 iterations of a particular key-strengthening function. It is a security parameter. Higher is stronger. 600,000 is a good number.

I opened the file. The shipped code reads 100,000.

A six-fold gap between what the document said and what the software did. No one lied. At some point the number in the code changed and the number in the document did not, and everybody involved continued to believe a thing that had quietly stopped being true.

It was found by reading one line of each.

I am telling you because a chapter about checking things, written by somebody pretending he had checked everything, would be worth nothing to you.

Here is why it matters more than it looks.

To whoever finds that gap, it doesn't read as documentation rot. It reads as a lie. They do not have access to the meeting where nobody decided anything. They have a document that says one thing and a system that does another, with your name on the document.

And this is where the stakes stop being about accuracy at all.

### Why the machine's mistake costs you more

You would be far more forgiving of a person than you'll ever be of a machine.

A colleague gets a number wrong in a deck and you think: bad week. Same number, same deck, and it emerges the machine produced it and nobody checked, and it doesn't read as a bad week. It reads as carelessness with a system you did not respect enough to supervise.

I don't know that this is fair. I know that it is true, and true is what you have to plan around.

So understand what you're actually protecting, because it's not the accuracy of a document.

**It is treated as something you claimed. As a lie. And other people find out.**

The loss of trust from one wrong figure is enormous and it is not proportionate to the size of the error. It attaches to everything else you've given them, including the parts that were right. People don't go back and re-audit your previous work when they find one bad number. They just quietly discount all of it.

Which means the thing you're building, when you build the habit of checking, is not a reputation for accuracy. It is a reputation for being someone whose work does not need checking. That is a much rarer thing and it's worth a great deal more.

And this is the difference. Not between someone who uses these tools and someone who does not, because soon that will be everybody. **The difference is between a professional who can be trusted with this and someone who just uses it.** One of those gets paid.

### Asking again is not free

I owe you the other side, because everything above pushes in one direction and a technique with no cost is a technique somebody is overselling.

We had a line in our outreach that worked. It closed messages warmly, it tested better than anything else, and the internal write-up celebrated it as the thing that made people reply.

The line was *either way, rooting for you*.

In Australian English, *rooting* is crude slang. It isn't a subtle regionalism. It is the kind of word that ends a conversation with somebody you were trying to build a relationship with.

It went out. In a real message, to a real person, in April.

It had been caught. Four separate checks in our system knew about that word. The one check sitting at the point of sending did not, and that is the only one that had to fail.

So: the version that tests best can be the version that ends the relationship. Iterating towards what performs will find you things that perform, and performance isn't the only thing you're optimising for. Somewhere in the loop, something has to be asking a question that has nothing to do with whether it works.

There is a second cost and it is quieter. If you check everything, you check nothing, because you won't sustain it and you'll start waving things through to catch up. The signal is what makes this affordable. You are not checking all your work. You are checking the parts where the sentence turned up in your head, and that's a small number of things per week.

Which brings me back to the case I left open.

### When the feeling is wrong

Sometimes I get it, run the checks, and the first answer was right all along.

Six minutes, no error found, nothing changed.

That is not a failed check and I want to be precise about why, because it is the most useful idea in this chapter.

What you bought was not a correction. What you bought was the **right to put your name on it.** Before the check you were about to send something you privately hoped was right. After it, you're sending something you know is right. Those are different objects, and only one of them lets you stand behind it when somebody pushes back in a meeting.

There is no version of this where you get that feeling and are better off ignoring it. Either you find something, and you've saved yourself. Or you find nothing, and you've converted a hope into a certainty for the price of six minutes.

The only losing move is the one everybody makes, which is to feel it and carry on.

So start there. Not with a system, not with a process, not with anything you have to remember. Just stop overriding the sentence in your own head.

Everything after this chapter is about what to do once you've stopped, and the first of those things is the one I've been circling since the tenement number.

Because noticing that an answer might be wrong isn't the same as being able to make it prove that it's right, and the second one is a thing you have to build.


## 7 · Make It Prove Itself

*Are you sure?*

Yes.

Of course it said yes. You asked the same system the same question in a slightly more anxious tone, and you gave it every signal that you wanted reassurance. It gave you reassurance. That isn't a check. That is a conversation with a mirror.

The last chapter was about noticing. This one is about what you do about it, and the honest news is that the obvious move doesn't work. *Double-check that* is worthless too, and so is *is this accurate?* All three produce the same output: a confident restatement of what it already told you, sometimes with a hedge bolted onto the front so it sounds humble.

What works is stranger and it takes about twenty seconds longer.

**You have to give it a reason to disagree with you.**

### The mistake everyone makes first

Picture what you actually do; you have an answer you half-trust. You paste it back in and ask it to check its work.

Think about what you have just handed it. The conclusion, the reasoning that produced the conclusion, and your evident hope that the conclusion survives. It now has three pieces of information, and all three point the same way.

There is a word for asking someone to grade work while showing them the answer they already gave and looking hopeful. It isn't a check. It is a formality.

The whole of this chapter is variations on one move: **change the thing's incentives before you ask it anything.**

### Make it solve the problem before it reads yours

Start here, because it's the simplest version and it works immediately.

We built a study tool for a teenager sitting exams. One system produces an answer to a question. A second system marks it. That much is obvious.

The instruction to the marker is the part that matters, and it's one sentence:

> *First solve the question yourself, from scratch, before you read the proposed answer properly. Do not anchor on it.*

That is it. That is the whole technique.

Without it, a marker reads the answer, finds it plausible, and confirms it. Plausible is a very low bar and almost everything wrong is plausible, otherwise nobody would have written it down. With it, the marker arrives at the problem with its own answer already formed, and now there are two answers to compare rather than one answer to approve.

You can do this on anything. A forecast, a summary, a recommendation, a price. *Work this out yourself first. Then read what I have and tell me where we differ.*

The difference in output is not subtle. You stop getting *this looks correct* and start getting *I get a different number, here is where it diverges.*

Here is the whole thing on an ordinary piece of work, because it is thirty seconds of typing and almost nobody has seen the two sitting next to each other.

Picture a cost estimate for moving a team of twelve into a new office · invented for this example, so treat every figure in it as illustration rather than as one of mine. Twelve desks, the fit-out, cabling, a month of overlap on both leases, and the removal itself. The spreadsheet says £84,000 and it is going to the finance director on Thursday.

You already know what happens if you paste it in and ask whether it looks right. The structure is logical, the major categories are covered, the contingency seems appropriate, and you might want to confirm the cabling quote. Courteous, accurate, useless · because it read your arithmetic before it did any of its own, and from that moment there was only ever one answer available to it.

So do not send it. Send this instead:

> *I am moving twelve people into a new office. Fit-out, desks, cabling, one month of overlap on both leases, and the removal. Before you read anything of mine, build the estimate yourself from scratch and show your assumptions. Then I will send you mine and you tell me where we differ and which of us is wrong.*

It comes back at £108,000, and the interesting part is not the gap. It is the four lines underneath it, because two of them are things you had thought about and two of them are not.

It has assumed dilapidations on the old lease. You had not, because your landlord has never mentioned it and you have not read the clause. It has assumed a week of reduced productivity across twelve people and priced it, which you would probably argue with, and you should · but you will argue with it out loud, in front of the finance director, which is a much better position than being asked about it and having nothing.

**And then the one that pays for the whole exercise.** It has your overlap at one month and flagged that a fit-out of this size slipping by two weeks is the ordinary case rather than the bad case, so the overlap is the line most likely to move and the one you have least control over.

That is not a checking result. That is somebody thinking about your problem before they thought about your answer, and you got it by withholding your answer for ninety seconds.

Now send yours and ask where you differ. What comes back is a list you can act on, and the £84,000 either survives with reasons attached or it does not, and both of those are better Thursdays than the one where a courteous system told you the structure was logical.

And when they agree, that agreement means something, because it was arrived at twice from two directions rather than once and then nodded at.

### Give it a different job, not a different tone

The second move is to stop asking it to check and start telling it who to be.

There is a piece of our system whose entire job is to attack a proposal before it goes anywhere; it is cast as a diligence analyst at a major bank looking at an acquisition. The instruction includes, in plain words:

> *Your job is NOT to be a cheerleader.*

And it isn't advisory. If it returns anything marked critical, or more than two marked high, the thing doesn't proceed. The gate is code, not judgement.

Notice what has changed. Nobody asked it to be more careful. Careful is a temperature setting and it does nothing. It was given a **role with different incentives**, and roles carry behaviour that instructions don't.

A diligence analyst who approves everything is bad at their job. That is the load-bearing part. The role itself supplies the motivation to find something, so you no longer have to keep asking.

You can do this in one sentence with no system at all. *You are the person in the room whose job is to stop this going out. What stops it?*

Or, from the last chapter, four words that carry the whole posture: **use an adversarial agent.** It knows what that means.

There is a smaller version of the same idea running elsewhere in our estate, checking outbound writing for anything that could be read as defamatory. It is cast as a cautious media lawyer, and the reason it is cast that way is written into the code as a comment:

> *A false positive costs 3 seconds. A false negative costs a relationship.*

That is the calculation to make before you decide whether a check is worth running. Not *how likely is this to be wrong.* **What does each kind of wrong cost me?** When those two numbers are that far apart, you do not need the check to be right very often for it to pay.

### Starve the checker

Now the least obvious idea in this chapter, and the one I would most like you to take.

**Give the checker less information than you gave the worker.**

Every instinct says the opposite. More context, better judgement — that is true when somebody is doing the work. It is false when someone is checking it, because most of that context is the reasoning you are trying to test, and handing it over is handing over the answer.

We do this deliberately in four places and each one deprives the checker of something specific.

The exam marker is not allowed to read the proposed answer before forming its own.

A quality reviewer that assesses whether a piece of client correspondence did its job is given **only the original inbound message**. Not the brief, not the strategy, not what we were trying to achieve. It has to work out what the job was from the client's own words. If it can't, we didn't understand the client.

A confidence challenger, whose job is to push back on how certain a claim is, is given the claim and a count of how many sources support it, and **never the reasoning that produced the confidence**. Its instruction says so directly:

> *You are NOT the analyst who made this claim. Your incentives are opposite: they want high confidence, you want accurate confidence.*

And a set of assessment lenses is deliberately run blind, without knowing whose interests they're supposed to be serving, so that the assessment isn't quietly bent towards the answer somebody wanted.

The pattern is the same every time. Whatever would let the checker shortcut to your conclusion is exactly the thing you withhold.

### And the counter-example, which is why this is a chapter

Now the part that stops this being a neat rule you apply everywhere and get burned by.

We took the same idea and applied it in the wrong place.

We had a grader, whose job was to score drafts on how well they read the situation. And in the spirit of blindness, it was run **without the original message the draft was responding to.**

The results are recorded. In seven of twenty audited items, the grader produced some version of the same sentence:

> *Cannot assess decode accuracy without seeing the original inbound.*

It couldn't do the job. It said so. And because it still had to return a number, it returned low ones, capping the average at **31.9 out of 100.**

Every one of those scores was garbage, and they looked exactly like real scores. A number arrived, it was in range, it went into the average. Nothing anywhere flagged that a third of them were the machine telling us it had been given an impossible task.

Here is the distinction, and it's worth memorising because it is the difference between the technique working and the technique poisoning your data.

**Starve the checker. Never starve the grader.**

A checker asks *is this right,* and it needs to be prevented from seeing your answer.

A grader asks *how good is this,* and it needs the standard to measure against. Take the standard away and it doesn't refuse. It guesses, and hands you a number that looks identical to a real one.

Blindness is a tool. It isn't a virtue. Ask yourself which of the two you're building, every time, and if the answer is *both* then you need two of them.

### When you have only got one of them

Everything above assumes you can reach a second system, and the redaction story in the last chapter is the argument for why you should. I am not going to run those numbers past you twice.

Here is the question that chapter left open. It is Tuesday, you have one tool, your employer has approved exactly that one, and a second opinion from a different company is not available to you today.

**Then change the role instead of the model.**

It is the weaker version and I want to be straight about how much weaker. A different model brings a different set of assumptions, and assumptions are the thing you are actually testing. A different role brings the same assumptions with a different attitude on top of them. It will not catch what the system simply does not know.

What it does catch is the far commoner failure, which is a system agreeing with you because agreeing was the path of least resistance. **A diligence analyst who approves everything is bad at their job. A helpful assistant who approves everything is doing exactly what it was asked.** Same engine, opposite incentive, and the incentive is most of what you were buying.

So use the role when the model is all you have, and know what you have bought: protection against easy agreement, not against a shared blind spot. The second is why chapter nine exists.

### The three things to type

Everything above collapses into three instructions. None needs any tooling.

**1.** *Work this out yourself before you read my version. Then tell me where we differ.*

**2.** *You are the person whose job is to stop this. What stops it? List every reason this fails, worst first.*

**3.** *Argue the opposite of your own conclusion, as convincingly as you can. Then tell me which argument is stronger and why.*

That third one is worth a note. We run a version of it on a system that produces group assessments, and it measures how much the voices agree with each other. When agreement gets too tight, it does something deliberate: it takes the most analytical of the voices and instructs it to

> *argue the OPPOSITE case regardless of your personal assessment.*

Because unanimity isn't evidence. Unanimity is very often the sound of everyone reading the same thing and finding it plausible. A room that always agrees isn't a room that has checked anything, and this is as true of people as it is of software.

### Confidence is not evidence, and the two are not even related

Now the most important idea in this chapter, and it is the one that will save you from the worst mistake available to you.

These systems produce confidence and accuracy almost independently. Not loosely coupled. **Almost independently**, and the gap is far too wide to lean on.

The clearest demonstration I know is a published result that our own confidence gate cites in its code, as the reason it exists at all. On a reasoning benchmark, a system reported an average of **89 per cent confidence** across five problems.

It got **zero of the five** right.

Read those two numbers together until they sit properly. Not one wrong out of five. Not three. All five, at eighty-nine per cent confidence, with no internal signal whatsoever that anything had gone wrong.

There is no mechanism in there that connects how sure it sounds to how right it's. It has never had one. Confidence is a writing style.

So we stopped asking it how confident it was and started counting instead.

The rule is arithmetic and it overrules the model's own judgement no matter how convincing the argument:

| Distinct sources | Maximum confidence allowed |
|---|---|
| None | 35% |
| One or two | 50% |
| Three or four | 75% |
| Five or more | 95% |

If it wants to claim ninety per cent on something with no sources behind it, it does not get to. The ceiling is thirty-five and the ceiling is code.

You do not need our system to use this. **Count the sources yourself.** It takes ten seconds. Zero sources isn't a weak claim, it is a different kind of object, and treating it as a fact you can act on is how people end up standing in front of a room defending something that was never checked.

There is one more piece of that gate worth stealing. A second system is brought in specifically to argue the confidence down, and when its verdict differs from the original, **the original is kept alongside it rather than overwritten.** You can see both. If the challenger inflates confidence by more than twenty points, the whole thing is rejected outright, because a challenger that talks you up isn't a challenger.

Keep the first answer. The disagreement is the useful part.

### So I went and looked

I have described a set of mechanisms for making things prove themselves. They are real, they're running, and I have shown you the code comments.

Then I did to them what this chapter has spent twenty pages telling you to do to everything else. I stopped asking whether they looked right and went and counted.

Here is what came back, and it is worse than not knowing.

Start with the simple part. The place where a challenge result is supposed to accumulate, so that I could look back and say *the challenger overturned the original answer in X per cent of cases*, holds zero records. Not thin data. None. I checked every copy of it I could find across our systems, in case the results had landed somewhere I had forgotten about, and they had not.

My first assumption was that somebody had forgotten to switch on the logging. That would have been the comfortable answer and it is not the one.

**The challenger has never run.** One of the two things that triggers it is a flag, and nothing in the system has ever set that flag, so the queue it reads has been empty since the day it was written. It has been sitting there the whole time, correct, complete, and waiting for work that could not arrive.

Then I looked at the live path, which does not depend on that flag, and this is the part I would like you to hold on to.

It fires when two conditions are met at once. Agreement has to be tight, and the spread of opinion has to be narrow. Two safeguards, and I had read that line more than once and thought it looked careful.

**They are the same condition.** Both of them measure how far apart the votes are, one as a spread and one as a statistic derived from the spread, and the first one is written so loosely that it was satisfied on **fifty-eight occasions out of fifty-eight**. It has never once been false. So a gate I believed was two independent tests is one test wearing two coats, and it took me four minutes with a calculator to find that out, having never once thought to look.

And one more, because it is the detail that stops this being abstract. The condition that does the actual work triggers when the spread is fifteen or below. **Fifty-two of those fifty-eight sit at seventeen or eighteen.** The entire working population of this thing is clustered two or three points above the line at which it does anything at all.

So the honest position on this chapter is not that I built an adversary and forgot to measure it. It is that I built an adversary, wired it in, admired it, and it could not have fired. Every case I have shown you above is real and every one of them was produced by a different mechanism than the one I was proudest of.

The repair is specified and handed to the people who own that code, with the arithmetic attached so nobody has to find it twice.

There is something almost funny about that, and it is worth sitting with rather than laughing at. The easiest system to forget to check is the one whose whole job is checking. It looks like diligence. It reports success. Nobody audits the auditor, because the auditor is the thing that was supposed to stop this happening.

### What this is actually for

Step back from the mechanics, because the mechanics aren't the point and I don't want you to leave this chapter with three prompts and nothing else.

What you're building isn't accuracy. Accuracy is a by-product.

You are building the ability to say, out loud, in a room where somebody is pushing back: *I know this is right, and here is how I know.* Not *the system said so.* Not *I checked it.* **Here is the thing I did that would have caught it if it were wrong, and it did not catch anything.**

That sentence is rare. Most people can't say it about their own work, and they can't say it because they've never built anything that could have caught them.

And you have just watched what it costs to be able to say it honestly. I can tell you the tenement figure is right, and I can name the check that found it. I cannot tell you the adversarial gate has ever caught anything, because I went and counted and it has not. **Both of those sentences come out of the same habit**, and you do not get to say the first one unless you are willing to say the second.

That is the whole trade. Somebody who only tells you what worked is telling you about half of a process, and you have no way of knowing which half.

It is also the sentence that separates the two people I keep coming back to. One of them uses these tools and produces work that's usually fine. The other one can stand behind it. Only one of those is being paid for judgement, and judgement is the only thing in this whole field that's not getting cheaper.

So make it argue with you. Give it a reason to. Withhold what would let it agree with you cheaply, and never withhold what it needs to be fair.

And then, when it has fought you and lost, you'll have something you didn't have before, which is a number you would put your name on.

Which raises the question I've been avoiding for two chapters, because a number you would put your name on isn't the same as a number that's good enough, and nobody has ever told you where that line is.


## 8 · The End of Good Enough

**Fifty-one out of a hundred.**

That was the score on a piece of writing I was about to send. I had read it twice. It was fine. It was clear, it was polite, it said what it needed to say, and if you had shown it to me in a meeting I would have nodded and moved on.

Fifty-one.

The reason came with it, and it is nine words long:

> *The email reads like it could be sent to anyone.*

I want you to notice what happened there, because it's the entire chapter.

I couldn't argue with that. Not because the machine has authority, it doesn't. Because the sentence was **specific enough to check.** I read the thing again with that one accusation in my head and it was true, obviously true, true in a way I had read past twice.

*Fine* isn't a standard. *Fine* is what work looks like when nobody has measured it.

### The thing you have never been told

Nobody has ever told you where the line is.

Think about that. You have produced work for years. Somebody occasionally said it was good, somebody occasionally sent it back, and in between you developed an instinct for *good enough* that has never once been calibrated against anything.

That instinct is mostly a measure of how much time you had.

And it's the single biggest cap on what you produce, because you stop when it feels done, and it feels done long before it's finished. Not through laziness. Through having no signal that tells you otherwise.

You can now have that signal, in the time it takes to read a paragraph, on anything you make. Almost nobody asks for it.

### Ask for the number

The instruction is this simple.

> *Score this out of 100 on five dimensions: hook, audience, proof, emotion, call to action. For each one give the score, one sentence on why, and one fix. Then tell me how many points each fix would gain, and rank them highest first.*

That is it. That is the technique. What comes back isn't a compliment or a criticism, it is a **structure you can act on** · the difference between *this could be stronger* and *your opening assumes they remember the last conversation, and they don't.*

Four things in that instruction are doing the work.

**Named dimensions**, because one overall score tells you nothing about where to spend the next ten minutes. Mine are the five above and I have reused them for a year:

**Hook**. Does the first line earn the second.
**Audience**. Is this written for the actual reader or for a general one.
**Proof**. Is there evidence, and is it near the claim.
**Emotion**. Does it acknowledge what they're worried about.
**Call to action**. Is the next step obvious and easy.

Use different ones if your work needs them, but choose ones that mean something. *Clarity* and *quality* aren't dimensions, they're moods, and a score against a mood is a mood with a number on it.

**A reason per dimension**, because a number without a reason is just an opinion with a decimal point on it. If it scores your opening 60 and can't say why, the 60 is noise.

**One fix, not three.** Ask for three and you get a wish list. Ask for one and you get the thing that actually moves it.

**And a price on every fix**, which is what turns feedback into triage. *This could be stronger* is advice, and advice is a thing you nod at. *Fix the opening and you gain fourteen points, tighten the close and you gain three* is a decision about where the next ten minutes go, made before you have spent any of them.

One more, worth stealing for anything that arrives as a list: **make every row end in a verb.** A scored list is information. A scored list where each row says what to do about it · `fix now`, `fix if time`, `note it`, `leave it` · is a decision already made, and you are executing a queue somebody sorted rather than reading a report and deciding. Options, risks, candidates, suppliers, findings. **Never accept a ranked list without a verb on every row**, because a ranked list still leaves the whole decision with you, which is the work you were trying to hand over.

### What comes back, in full

I have described this without showing you one. The 51 at the top of this chapter came back as a single number and one sentence, which is what you get when you ask for a score and nothing else, and it is why the rest of this chapter exists.

So here is the fuller thing, on a different piece of work · a follow-up email to somebody who turned down a proposal six months ago and is being approached again. Run against five named dimensions, with normal defined.

> | Dimension | Score | Why | One fix | Points |
> |---|---|---|---|---|
> | Hook | **42** | The first line explains who you are. The reader knows who you are. | Open on the thing that changed since you last spoke. | **+14** |
> | Audience | **51** | Written for a general commercial reader, not for someone who has already turned this down once. | Name the objection they raised last time in the second line. | +9 |
> | Proof | **68** | The claim about turnaround sits three paragraphs from the only number supporting it. | Move the number next to the claim. | +6 |
> | Emotion | **44** | Does not acknowledge that the last attempt failed and they are being asked again. | One sentence naming it before the ask. | +11 |
> | Call to action | **60** | Asks them to "let me know your thoughts". | Replace with one dated, binary ask. | +8 |
>
> **Overall 51.** Ranked: Hook +14, Emotion +11, Audience +9, Call to action +8, Proof +6.

Look at what that is and is not. It is not a critique, and it is not encouragement. It is a queue, sorted, with the arithmetic already done, and the top line is worth nearly three times the bottom one.

**Now watch what most people would have guessed.** Ask anybody which line of that email is weakest and they say the call to action, because it is the visible one, the one everybody has been told about, and the one they know how to fix. It scores 60 and it is worth eight points. The opening is worth fourteen, and the opening is the part that had already been read twice and approved.

That is the pattern, and it is why the ranking column earns its place. **You are drawn to the problems you already know how to solve**, and your instinct about which problem matters is wrong in a direction you cannot catch from the inside, because the problems you cannot see do not announce themselves as problems. They read as finished.

Do the top-ranked fix. Do not do the list · the list is how you spend an hour and gain twenty points that a re-score will not give you credit for, because you will have broken something else on the way through.

And then the part that's genuinely uncomfortable: you have to be willing to act on a bad score for work you already liked.

### Tell it what normal looks like, or everything scores 91

There is a problem with everything I have just told you, and it is the reason the scoring habit fails for most people who try it.

Ask a thing to score its own work and it drifts high. Most of the time it hands you ninety-one. The number is sincere and it's worthless, because nothing has told it what the range means.

The fifty-one at the top of this chapter is the exception, and the exception is why I went looking. An uncalibrated score that comes back low is worth paying attention to precisely because the pull is the other way.

**The fix is one sentence and it is the most valuable clause in this chapter.**

> *95 and above is exceptional and rare. Most first drafts land between 76 and 88. Score accordingly.*

That is calibration. You have supplied the distribution, so the score now sits inside a scale that exists rather than floating in one it invented. A 79 means something. A 91 has to be earned against a stated bar rather than awarded out of politeness.

Use whatever numbers match your world. The specific figures matter far less than the fact that you gave it any, and the difference in what comes back is not subtle — the same piece of work that scored 91 unprompted has, in my experience, come back in the high seventies once normal was defined.

That number in the high seventies is the useful one. It is the one that was scored against a stated bar.

**Then close the loop from score to action**, because a low score you have to interpret is a task you will not do.

> *For any dimension scoring under 82, give me one question that would fix it. Maximum twenty words. No preamble. One ask.*

Not a paragraph of advice. **One question, under twenty words**, which you can answer in thirty seconds and act on immediately. That constraint is doing the work. Given room, it will hand you a considered essay about your opening, and you'll read it, agree with it, and change nothing.

### Set the bar before you look

Here is the trap, and everyone falls into it once.

You score something, it comes back 62, and you think: well, 62 is not bad for this kind of thing.

You just moved the bar. You moved it after seeing the score, which means the bar is now whatever your work happens to be, which means you have built an elaborate mechanism for confirming that everything you do is acceptable.

**Decide the threshold before you see the number.** Write it down if you have to. And set it **above** the top of the normal range you just described. If ordinary first drafts land in the eighties, a bar of 75 passes everything and measures nothing. And a bar of 88 is no better, because 88 is the top of the range you just called normal. *This doesn't go out below 92.*

We do this structurally, and I want to give you the rule before the numbers, because the numbers are only the rule made checkable.

The rule is one sentence and I have been saying it to people for years. **We only write to somebody if there is a value add · if we are serving the person receiving it, telling them things they did not know, and going past what they expected of us.**

That is easy to agree with and impossible to enforce, because on a Thursday afternoon everything you write feels like it serves the client. So it became three tests, and nothing goes to a client until it passes all three.

**Does it serve them?** Above 75.
**Does it serve us?** Below 20. A message that is mostly working for you is an advertisement, and the person opening it can feel that in the first line whether or not they could name it.
**Does it tell them something they did not already know?** Above 60. If they finish it no better informed than they started, we have taken their time and given them a feeling.

All three. Any one fails and the thing does not go, no matter how well written it is · which is the whole point, because well written is exactly how a self-serving message gets sent.

Read the middle one again, because it is the one nobody builds. Everybody has some version of *is this good*. Almost nobody has *is this actually for them*, and that is the test that separates a firm people stay with from one they tolerate.

There were four for over a year. The fourth required the serving score to beat the self-interest score by at least 40. It is still in the file. It has never once rejected anything, and it never could have: the first gate puts serving above 75 and the second puts self-interest below 20, so the gap is never less than 55, and 55 always clears 40. A gate that can't fail isn't a standard. It is a decoration that makes a list look more rigorous than it's, and it survived a year of use because nobody applied the question this chapter is about to the mechanism doing the checking.

There is a smaller structural version of the same idea. For our most sensitive category of outbound writing, the system is capped at 60 and cannot score itself higher. Not because that work is worse. Because a system that can award itself top marks on its most consequential output will, eventually, do exactly that.

Count your own gates. Then work out which of them has ever actually stopped anything, and which of them is about the reader rather than about you.

The instruction that sits above those gates is one line, and it's the whole posture of this chapter:

> *HONEST SELF-ASSESSMENT: reject your own work if it's mediocre.*

You won't do this reliably by intention. Intention is what fails at 6pm on a Thursday. You do it by writing the threshold down first, when you're calm and nothing is at stake, and then treating it as though somebody else set it.

One thing this chapter deliberately does not cover, because it belongs next door. How much to trust the number the tool gives you about its own confidence is chapter 7's problem, not this one. Here we're setting your bar, not testing its certainty.

### Somebody has to be the ruler

A scoring system that grades itself against its own opinion of good is a closed loop, and closed loops drift somewhere pleasant and stay there.

So one thing in ours isn't negotiable. Every day, a job sends me three variants of a piece of writing. I reply with **three integers between 0 and 100.** That is the whole interaction. Thirty seconds.

Those numbers go into a file that shapes how everything downstream is weighted, and the rule sitting at the top of that file is written in blunt language on purpose:

> *GROUND TRUTH RULE: John's scores are the truth. Not Claude's opinion.*

If the reply can't be parsed, it asks again. It never silently drops a score, because a missing score that quietly becomes an average is worse than no score at all.

Thirty seconds a day, at exactly one point in the process, and that point is the scoring point. Everything else runs without me.

That is the shape to aim for in your own work. You aren't trying to check everything. You are trying to be the ruler that the measuring is calibrated against, and to be that in as few minutes as possible.

### What the number is really for

I have given you a technique. Now the reason it matters, which is not accuracy.

A number is something you can hand to another person.

*I think this is good* is not transferable. It ends the conversation or it starts an argument about taste, and in either case the most confident person in the room wins.

*This scored 51, and the reason was that it reads like it could be sent to anyone* is a completely different object. Your colleague can disagree with the dimension. They can disagree with the weighting. They can say the score is wrong and explain why. **All of that is progress**, and none of it is available when the alternative on the table is somebody's feeling.

This is what your senior people actually want and rarely get. Not certainty. Something they can push on. A recommendation that arrives with its own scoring, its alternatives, and what those alternatives scored, hands them the ability to interrogate it in ninety seconds instead of asking you to go away and think again.

That is the difference between being someone who submits work and being someone who is trusted with decisions. And it isn't a personality trait. It is a habit of attaching numbers to things.

### The way this goes wrong

There is one way to do all of the above and get nothing from it, and it is common enough to be worth naming.

You start scoring things. You enjoy it. Six months later there is a drawer full of numbers and nobody has ever checked whether the high-scoring work actually did better than the low-scoring work.

**A scoring habit that has never been tested against an outcome is a ritual.** It feels like rigour and it is decoration.

The fix costs twenty seconds a time. For one month, write down the score and write down what happened · the reply, the decision, the silence. That single column is the difference between measuring something and performing measurement, and almost nobody keeps it.

### Start tomorrow

One thing, on the next piece of work you were about to call finished. Score it before you send it.

The first time you do this you'll find something. Everybody does. And the version of you that ships the fixed one isn't the same professional as the version who shipped the first one, which is what the last chapter of this book is going to be about.

But there is a step before that, and it's the one people skip because it looks like extra work rather than the thing that makes all of this affordable.

Because so far I've had you asking one system, checking one system, scoring one system. And the whole time, the person sitting next to you has been getting worse answers for more money by doing exactly that.

---

> ### ▪ DO THIS
>
> **Score the next thing you were about to send, before you send it.**
>
> **1. Write your threshold down first**, and put it **above** your normal range rather than inside it. *This does not go out below 92.* A bar set inside the range passes everything and measures nothing.
>
> **2. Paste this under your work.**
>
> > *Score this out of 100 on five dimensions: hook, audience, proof, emotion, call to action.*
> >
> > *95 and above is exceptional and rare. Most first drafts land between 76 and 88. Score accordingly.*
> >
> > *For each dimension give the score, one sentence on why, and one fix. Tell me how many points each fix would gain and rank them highest first.*
> >
> > *For any dimension under 82, give me one question that would fix it. Maximum twenty words. No preamble.*
>
> **3. Do only the top-ranked fix.** Not the list. The one worth the most points.
>
> **4. Re-score.** If it meets your threshold, send it. If not, do the next fix. And stop after two. A third pass on the same piece is the too-deep failure.
>
> **Takes four minutes.**
>
> **When it scores below your bar and the deadline will not move:** send it, write the score down, and fix the pattern next time. A rule that only works on unhurried days is not a rule.
>
> > **You will know it worked when** a piece of work you had already decided was fine comes back under your own threshold. And you fix it anyway.


## 9 · Use More Than One

Open a second tool. Put the same question in. Look only at where the two answers part company, and spend your attention there.

Ninety seconds, and that's the entire method. The rest of this chapter is why it works, and why the people who most need it are the ones least likely to do it.

Ask yourself a question about a hospital.

Does a patient on a ward need a surgeon to check on them? Yes. Obviously yes. There are decisions on that ward that only a surgeon can make, and getting them wrong isn't recoverable.

So does the surgeon do every check?

Of course not. Nurses check on that patient far more often than the surgeon ever will. They take the observations, they watch the trend, they know what normal looks like for this person today. And when something moves, they call the surgeon back in.

Nobody thinks this is a compromise. Nobody thinks the hospital is cutting corners. It's simply how you run a ward. A surgeon doing hourly observations on forty patients would be ruinously expensive and, by about hour six, worse at it than the nurses.

**That is how you run these tools, and almost nobody does it.**

The person sitting next to you has one. They use it for everything. The hard analysis and the tidy-up of a paragraph, the strategic call and the reformatting of a list. They are paying surgeon rates to take somebody's blood pressure, and they're getting worse answers for more money, because the expensive one isn't better at everything. Judgement is where it earns its money, and most work isn't judgement.

### What we actually do

Twenty-five different jobs run across our systems. Each one has a written policy about which tool handles it.

**Eighteen run on a cheap, fast model. Six are pinned to the expensive one. Which leaves one I cannot place from my own table, and I have left it there rather than tidy the count.**

The ratio isn't the interesting part. The interesting part is that every single pinning carries a **written reason**, in plain language, sitting in the file next to the setting. Two of them read:

> *IP-sensitive outreach, quality non-negotiable.*

> *Acquisition decisions, Opus tier required.*

There's the discipline, and you can adopt it this afternoon without any of our software.

**If you cannot write down why this particular job needs the expensive one, it does not.**

Try it on your own week. Most of what you do won't survive that sentence, and the things that do survive it are the things you should be spending your attention on anyway. The exercise is not really about cost. It is a way of finding out which of your work is judgement and which of it's typing, and most people have never separated the two.

### The nurses report back

The part of the analogy that does the real work is the last part, and it's easy to skip.

The nurses don't just do the cheap tasks. **They report back.** They gather the observations, they notice when something is outside the normal range, and they escalate. The surgeon's expensive attention is spent on the escalation, not on the gathering.

Build that shape. Cheap and fast for collecting, listing, extracting, formatting, first passes, checking things against a rule. Expensive for the judgement call that arrives at the end of all that, holding everything the cheap ones brought back.

A researcher gathering forty documents doesn't need to be brilliant. A person deciding what those forty documents mean does.

Set it up the other way round, which is what happens by default when you use one tool for everything, and you spend your best resource on the gathering and arrive at the decision with nothing left.

### Two of them disagreeing is the point

Now the other half of using more than one, and it has nothing to do with cost.

**One system agreeing with itself is a single point of failure with a confident voice.**

You saw this in chapter 6 from one direction: a checker that has seen your answer will confirm it. Here is the other direction. Two genuinely different systems, given the same problem, produce genuinely different answers, and **the gap between them is information you cannot get any other way.**

When they agree, you have something. Not proof; they aren't independent, they read much of the same internet and fail in correlated ways. But two routes that don't share your framing beat one route travelled twice.

When they disagree, you've found the thing worth your attention, and you have found it in about ninety seconds. That disagreement isn't a problem to resolve before you can get on with the work. It **is** the work. It is the map pointing at the one paragraph where the difficulty actually lives.

Which is what the previous two chapters were driving at when they kept saying a different **model**, not just a different question. Rephrasing gets you the same assumptions back. Only a different system gets you different ones, and assumptions are the thing under test.

Nothing about this requires infrastructure. Open a second tool. Paste the same question. Look at where the two answers part company, and spend your six minutes there.

### Which two

I have written a chapter called *Use More Than One* and not yet told you which ones I use, and you would be entitled to read that as evasion, because in most books it is.

So: as I write this, in the middle of 2026, I use **ChatGPT and Claude**.

Now the caveat, and it is the reason most books on this subject will not name anything at all. By the time you are holding this, at least one of those names will have moved. A version number will have changed, a capability will have arrived that reorders which one I reach for first, or one of them will have been renamed by a marketing department. That is not an argument for saying nothing. It is an argument for **dating the claim**, so that you can see how old it is and discount it yourself, which is exactly what chapter sixteen asks you to do to your own work.

My pair matters less than whether yours are genuinely two, and this is the practical trap. **A great many products that look like competitors are running the same engine underneath**, licensed from the same handful of companies and dressed by different marketing departments. Your bank's assistant, your CRM's assistant and the tool your team bought last year may all be one system with three logos, and comparing them will feel exactly like a second opinion while giving you none.

So find out what is actually underneath before you trust any comparison. It takes one search.

Then stop taking my word for the pairing. Put the same real question through whichever two you can reach, on a decision you have to make this week, and watch which pair disagrees in ways that help you. That costs an afternoon, it is the only test that answers this for your job rather than for mine, and you finish it having picked your tools on evidence instead of on somebody else's habit.

One caution, because a second tool is a second place your work now lives. Whatever your employer approved, it probably approved one system and not two. Chapter fourteen shows how to strip the part that matters before it leaves, which makes this a question of habit rather than permission.

### The reasons matter more than the ratio

I gave you eighteen and six. Do not copy the ratio.

Our eighteen and six comes from the shape of our work, and yours will be different. What transfers is that **every one of the twenty-five has a reason attached that a human wrote.**

None of that's bureaucracy. What it buys you is a decision you can review later. By you, once you've forgotten why you chose it, or by whoever inherits it. A setting with no reason attached is indistinguishable from an accident, and six months later nobody can tell whether it was a considered choice or a default nobody looked at.

Apply the same rule to your own habits. Always reaching for the expensive one on a certain kind of task? Write down the sentence explaining why. Can't? You've found something to change. Can? You now have a decision you can defend when somebody asks why the bill looks like that.

### Know when you paid without deciding to

A quieter failure, and one you won't notice for months.

Systems fall back. Something is unavailable, something times out, something is rate-limited, and quietly the work routes to the expensive path instead. The output is fine. Nothing is broken. Nothing tells you.

We found ours the ordinary way, which is that the numbers looked wrong. The note in the code afterwards is blunt:

> *Silent fallback was costing real dollars on calls designed to be free.*

Every fallback is now counted and classified, so a route designed to be cheap that has been quietly running expensive for a fortnight shows up as a number rather than as a surprise at the end of the month.

Your version of this is smaller and it's the same shape. **Know which tool actually answered.** Not which one you meant to ask. If you have a paid tier and a free one, or a fast setting and a slow one, check occasionally that what you are getting is what you chose. The gap between intended and actual is where money and quality both leak, and neither leak announces itself.

### The counter-example, because I have made this sound too clean

I have spent this chapter telling you to be disciplined about cost. Here is what happened when we were.

We put in a cost-control gate. Sensible thing. It capped daily spend, it capped individual tools, and one of those caps was set to five dollars.

The gate did its job. It blocked automated calls once the limit was reached.

And a feature people were actually using stopped working. Users saw a failure. They didn't see *this is temporarily unavailable to control costs*, they saw the thing break.

**Saving money broke the product.**

The lesson isn't that cost control is wrong. The lesson is that a limit with no visible failure mode is a trap, and the person who set it's never the person who discovers it.

So when you build any rule for yourself about which tool to use, ask what happens at the boundary. When the cheap one isn't good enough for this particular case, how do you find out? Answer comes back *the work is quietly worse and nobody says anything*? You've built the same trap in miniature.

The nurses call the surgeon. That is the part that makes the system safe, and it's the part everyone forgets to build.

### The thing you are depending on will be retired

One more, because it's the failure that catches people who have done everything else right.

On a day in June, a set of model identifiers we depended on were retired by the people who made them. Not deprecated with a long runway. Gone.

We had, by luck as much as foresight, a mapping layer that translated old names to current ones. The note recording it says that without it, **every single call across the entire system would have broken.**

Building your way of working around one tool from one company is an exposure, and that's what it looks like when it fires. They're not unreliable. The thing you tuned your habits to is a product, and products change underneath you without asking.

Using more than one isn't only about cost and it's not only about disagreement. And it's the reason a change at one vendor costs you an inconvenience rather than a week.

### Deliberately, not by accumulation

One qualification before you go and do any of this, because there is a version of it that costs you money and buys nothing.

Use more than one **deliberately**, for reasons you have written down. Do not end up with more than one **by accumulation**, which is what happens when nobody is watching · a tool arrives for one job, another arrives for a second, and three years later there are five doing overlapping work, not one of them ever chosen against the others, and all of them being paid for every month.

That is all of the cost and none of the benefit, because the benefit was never in the number. It was in two of them being genuinely different, and in somebody having decided which was which and why.

The difference between those two situations is whether anybody is holding the whole picture. That is your job now, and it is the part that doesn't get cheaper.

### What to do this week

Three things, none of which need any system.

**Sort one week of your own work** into judgement and typing. Be honest. Most of it's typing.

**Put the typing on the cheap fast tool** and see whether the output is worse. Usually it isn't, and the surprise is instructive.

**Take one decision that actually matters and run it through two different tools.** Read the gap. Not the answers, the gap.

That third one is where this chapter earns its place, and it takes about four minutes.

You will now have something no one else in your building has: a piece of work you can defend, produced for less, with the one contested paragraph already identified.

And every technique in the last five chapters has been about the work.

Not one of them has been about the person it's for. Which is the larger of the two problems, because a perfectly checked, well-scored, cheaply produced answer aimed at nobody in particular lands exactly the same way as no answer at all.

---

> ### ▪ DO THIS
>
> **Sort one week, then run one decision through two tools.**
>
> **1. Split last week's work into judgement and typing.** Be honest. Most of it is typing.
>
> **2. Put the typing on the cheap fast option** and see whether anything gets worse. Usually nothing does, and the surprise is the lesson.
>
> **3. Take one decision that actually matters** — not a task, a decision. And put the same question into two tools running on different underlying models. Check that they do. Plenty of branded assistants share one engine.
>
> **4. Ignore both answers. Find where they part company.** Write down that one point.
>
> **5. Spend your attention only there.**
>
> **Steps 3 to 5 take four minutes. Step 1 is a lunch break, once.**
>
> **When both answers agree:** that's worth something, but it isn't proof. Two routes to the same place beats one route travelled twice, and it's still two routes.
>
> **You'll know it worked when** the disagreement points at the exact paragraph you'd have got wrong. And you find it in ninety seconds instead of in the meeting.

---

So before you send another thing, there is a room to read. And almost nothing that matters in that room was written down anywhere in the message you are answering.


## 10 · Read the Room, Not Just the Message

You can answer a message perfectly and still lose the room. Nothing in the previous nine chapters will save you from it, because every one of them made you better at producing and none of them made you better at aiming.

A message arrives. Four lines, polite, asking for something specific.

You answer what it asks. You answer it well, quickly, and completely.

And it lands badly, or it lands nowhere, and you spend the next fortnight wondering what you got wrong, because on any reading of the words in front of you, you got nothing wrong at all.

You answered the message. You didn't read the room.

Aiming is where the value is, in a way that has nothing to do with the tools and everything to do with what you now have time to notice.

### Three layers, and you only see the top one

Under any request there are three things stacked, and the message contains one of them.

**What they said.** The literal ask, available to everybody.

**What they actually want.** Usually adjacent, occasionally the opposite. Somebody asking for a comprehensive comparison of five options frequently wants permission to choose the one they have already chosen. Somebody asking you to *review* something often wants it defended, not improved.

**Why they want it.** Almost never about the task. Wanting to look competent, wanting not to be blamed, wanting it settled so it stops occupying their week, wanting to be seen to have handled it by somebody two levels up who appears nowhere in the message.

*Can you send me the figures for last quarter* is a request for figures. Underneath it is usually *I have to defend something on Thursday*.

Send the figures and you have completed a task. Send them with the one line that anticipates Thursday and you have done something else, and the difference in how you are seen is nothing like proportional to the effort.

**The move is a question you ask yourself, not them.** *What will this person have to do next, and what will make that easier?* You do not have to be right about the third layer, only to have considered that it exists · at which point your output stops answering a task and starts answering a situation.

That is where the returned hour goes, and almost nobody spends it there.

### Whose urgency is it?

Every request arrives carrying a stated urgency. Every situation has a real one. **The two are frequently different, and the gap tells you almost everything.**

Stated far above real means somebody is manufacturing pressure · a supplier at the end of their quarter, a manager passing on a panic they never examined. **Do not let their urgency become yours.**

The reverse is rarer and more useful. Real above stated means somebody is under pressure they will not admit to: understated language, a favour that would take three days. Noticing that is worth more than replying quickest.

Two questions settle it. *What actually happens if this is late?* And *who is applying the pressure, and what is happening in their week?* If nobody can answer the first, that is a question to ask rather than a licence to slip. Ask it in writing, and let the answer set the date.

### Who else is in the room

The person who wrote to you is rarely the only person who matters, and they're quite often not the one who decides.

Before anything consequential goes out, spend a couple of minutes on the people who are not in the message.

Two things about each of them. **How much do they influence the outcome**, and **how likely are they to agree with where this is going.**

It is an old management grid and I didn't invent it. Two axes, in your head, thirty seconds.

The useful part is what falls out of it. Somebody with high influence who is likely to disagree is the single most important person in your work and they aren't on the email. Your output has to survive them. Not persuade them necessarily, but at minimum not hand them the easy objection.

Somebody with high influence who agrees is your route, and most people never think to use them.

And somebody with strong opinions and no influence will absorb an unlimited amount of your time if you let them, and the reason they get it is that they're the loudest voice in the thread.

The most common way careers stall on this is producing work that's correct, well-received by the person who asked, and quietly killed by somebody two rooms away who was never consulted and didn't have to be.

### The one instruction that changes the output

There is a practical version of all this and it's one line.

**Tell it who this is for.** Not the topic. The person.

And don't describe them in categories. Categories produce writing for an average, and the average reader doesn't exist.

The best version of this instruction I've used is one line:

> *You are writing for a board member scanning this on their phone at 6am.*

Look at what that does. It gives a **person**. Someone senior, busy, accountable. A **place** — a phone, which kills your three-line paragraphs before you write them. A **time**. 6am, before coffee, before the day has started going wrong. And a **constraint**. Scanning, not reading, so anything requiring a second pass is already lost.

Four facts, fourteen words, and every one of them changes the output.

Compare it to *write this for executives*, which is what almost everybody types, and which describes a demographic rather than a moment.

So build your own. Not *for the finance team*. **For the finance director who has already had this pitch twice and did not like it either time, reading between meetings.** Not *for a customer*. **For someone who has been let down by a supplier before and is looking for the reason to say no.**

A person, a place, a time, and one thing that's true of them right now: a worry, or a constraint like *scanning, not reading*. Write it once for the three people you produce work for most, and reuse it for a year.

This is the single highest-return sentence you can add to any instruction, and the reason it works is that without it, the thing is writing for an average of everybody, which is a person who doesn't exist.

And there's a diagnostic that costs nothing. When something comes back and feels subtly wrong and you can't say why, ask it **who it assumed the reader was.**

The answer is often the entire explanation. It has been writing for a general business audience while you needed something for one sceptical finance director who has heard this pitch twice before; you have found the fault in one question, without re-reading a word.

### Be them, do not imagine them

There is a stronger version of this and it turns on one word.

Most people, when they think about the other side, ask *what might they think about this?* That produces a polite paragraph of speculation, written from your chair, about a person who isn't in the room.

Change the instruction from *imagine* to *be*.

> *You are now [that person] receiving this. Answer as them, in their voice. Where do you push back, and what do you ask for that I have not given you?*

Not *what would she think*. **Be her and answer.**

The difference in what comes back isn't small, and it is the same difference you get from a good role-play in a training room versus a discussion about the role-play. One produces observations. The other produces an objection in her register, phrased as she might phrase it.

Here is the difference on one paragraph, and it is the clearest demonstration in this chapter.

You have written a recommendation to move three sites onto one supplier. It is well argued and the saving is real. You are sending it to a regional manager who has run her site her own way for nine years.

Ask what she might think and you will get a paragraph about concerns over local control, worries about service levels during transition, and a wish for existing relationships to be respected. Not wrong. Written from your chair, about a category of person, containing nothing you had not already suspected, and no use to you on a Tuesday.

Change one word. Not *imagine her*. **Be her:**

> *You are Angela, regional manager at the Northampton site for nine years. You are reading a proposal to consolidate your supplier into a national contract. Answer as her, in her voice. Where do you push back, and what do you ask for that I have not given you?*

> *Right. Who do I ring at four o'clock on a Friday when a delivery hasn't turned up? Because at the moment I ring Dave and Dave sorts it, and I've got nine years of Dave. Your paper says "national account management" and I've been on the end of national account management before · it means a ticket number and a callback on Monday.*
>
> *And what happens in the first six weeks? You've costed the saving from month one. I've done three of these and month one is always the worst month of the year. Who is covering that, me or you?*
>
> *One more. You've got the saving as a total across three sites. What's it worth on mine? Because if the answer is that Northampton gets slightly worse service so that two other sites get cheaper, say so and I'll live with it. Don't tell me it's better for everyone when I can see it isn't.*

Those are three specific things, in a register, that you can answer before she asks. The first one is a name and a phone number and it costs you an email to find out. The second is a line in the paper about the transition period. The third is the one that would have killed the whole thing in the meeting, and it would have killed it in front of other people.

Be precise about what that's worth, because it's easy to overclaim. **It is still a guess.** The machine has never met Angela. Every word of that is invented from a role and nine years and a consolidation, and the real Angela may care about none of it.

But you now have three questions to take to her, which is a completely different object from a paragraph of speculation. Its job is to hand you the question you take to her, not the answer you take into the room.

Do it for the person who decides, and then again for whoever loses something if this goes ahead. Because the second one is where the real resistance lives, and they are almost never the one you were writing to.

Then do the thing that makes it real: take the sharpest objection to an actual human and ask whether it lands. A guess you've tested is worth more than a guess you've rehearsed.

I use this on competitors as well. Not *how might they respond to this*, but **be each of them in turn and answer.** What they say back is frequently something I had genuinely not considered, and it arrives in about ninety seconds.

### The check that has nothing to do with the words

One more, and it's the one that catches the errors nothing else does.

Before anything goes out, put it in front of a different lens and ask a question that isn't about quality. Ask whether it **makes sense in that person's world.**

This is not proofreading and it isn't fact-checking. It is the question of whether a claim that's internally consistent is plausible in the reality the recipient actually lives in.

I have caught output claiming a sector was booming in a place where the situation on the ground made that impossible. Nothing in the text was wrong. Every sentence was defensible. It simply couldn't be true for anyone who knew the place, and sending it would have told that reader, in one line, that we didn't.

That is a category of error no amount of checking the work against itself will find, because the work is consistent; it is only wrong from outside.

So the last question before sending is not *is this right.* It is: **would the person receiving this recognise their own world in it?**

### Reading a room is not intuition

The word makes it sound like a gift somebody either has or hasn't, and that is the wrong word. **Intuition is what we call a process once we have forgotten the steps in it.**

Watch anybody who is genuinely good at this and they are not sensing anything. They are running a short list, fast, mostly without noticing they are doing it.

Who else will see this. What has already been decided. What the sender is afraid of. What happens to them if this turns out to be wrong.

Four things, every one of them knowable, not one of them mystical. And that matters to you for a specific reason: **a list can be learned and a gift cannot.** If reading a room really were intuition, you would either have it or you would not, and this chapter would be a waste of your afternoon.

A list gives you one other thing, which is a way of being wrong usefully. You cannot check whether your feeling about somebody was right. You can check, every time, whether you asked who else would see it.

### The part nobody measures

One warning before the list, because this is where the discipline usually dies.

People keep a record of what they produced and almost never of how it landed. The document is in a folder. Whether it worked is in nobody's file.

Which means the half of your work most dependent on judgement · the half that faces other people · is the half you have the least evidence about, and it stays that way for an entire career unless somebody decides otherwise.

You already have the fix from chapter eleven. One line before you send, on what you expect to happen. It costs eight seconds and it is the only way this ever becomes something you get better at rather than something you have opinions about.

### The three minutes

Before the next consequential thing you send.

*What will this person have to do next?*
*Who else decides, and would they object?*
*Is the urgency theirs, or is it real?*
*Would they recognise their own world in this?*

Four questions, about three minutes, no tools required.

Most people never ask any of them, which is why most work is technically correct and lands with a sound like nothing at all.

You now have the time to ask them. That is what the time you got back was for.

Which accounts for the machine, and the work, and the person receiving it.

It leaves exactly one participant in this arrangement unexamined, and it's the one who has been choosing every question in the book so far.

---

> ### ▪ DO THIS
>
> **Aim the next consequential thing you send at one named person, starting before you write a word of it.**
>
> **1. Write the reader in one line.** A person, a place, a time, and one thing true of them right now — a worry or a constraint. *For the finance director who has heard this twice, reading between meetings, worried it lands on her budget.* Not "for the finance team".
>
> **2. Put that line at the top of your instruction.** Before the task, not after.
>
> **3. Now write the draft. Then be them.** Paste it back with:
>
> > *You are now [that person] receiving this. Answer as them, in their voice. Where do you push back, and what do you ask for that I have not given you?*
>
> Then run it again as whoever loses something if this goes ahead. That one is where the resistance lives.
>
> **4. Answer their objection inside the draft**, before they raise it. That single move is what senior people notice. And if the objection matters, put it to the real person before the meeting rather than after.
>
> **5. Last check, and it is not about the words.** In a fresh window, with no history: *would a [their role] in [their situation] recognise their own world in this?* A thing can be internally perfect and impossible for anyone who actually knows the situation.
>
> **Steps 1 and 2 take ninety seconds.** Step 3 costs a round-trip and step 4 costs a rewrite. Budget ten minutes the first time and five after that.
>
> **When you cannot name the reader:** you have found the real problem, and no amount of rewriting will fix it. Go and ask who this is for.
>
> **The signal to watch for:** people acting on your work in the meeting rather than taking it away to think about. Treat that as the sign you aimed well, and start keeping a record of when it happens.


## 11 · Check Your Own Thinking

Ten chapters of checking the machine.

Brief it properly. Decide the depth before you ask. Ask a different question. Make it argue against itself. Score it. Use a second one. Read the room it is aimed at.

Every one of those checks runs on material **you** chose, framed the way **you** framed it, against a standard **you** set.

Which means there's one participant in this whole arrangement that nothing in the book has examined yet, and it's the one holding the pen.

Your errors survive every technique in the previous ten chapters. They survive because they are upstream of all of them; you can't catch a bad question by checking the answer harder.

### You ask from inside the first answer

Start with the one that costs the most, because it is the error these tools amplify hardest and it does not feel like an error at all.

I told you to ask a different question. What I did not tell you is that when you ask again, **you ask from inside the first answer's frame.**

The first response you read sets the terms. It decides what the categories are, what counts as a consideration, roughly what range the numbers live in. Every subsequent question you ask is shaped by it, including the ones you believe are challenging it, because challenging something is still accepting its terms.

You have felt this without naming it. Somebody gives you a figure and every estimate you make for the next hour clusters around that figure, including the ones you arrive at by reasoning you would swear was independent.

**Two defences, both cheap.**

Write down what you expect **before** you look. One line, thirty seconds, before the first answer arrives. Now you've your own anchor to compare against, rather than inheriting theirs.

And when the first answer was substantial, start a completely fresh conversation for the second opinion rather than continuing the existing one. Not because the tool remembers. Because **you** do, and in a fresh window your own questions come out different. Try it once and read both. The difference will bother you.

### The thirty seconds that shows you your own anchor

Here is what that one line actually catches.

You are working out what it would cost to bring a piece of work back in-house from an agency. The figures below are invented to show the shape. Before you ask anything, you write:

> *I think in-house lands around £60k a year and the agency is about £90k, so we save thirty.*

Then you ask properly, and back comes a fully-loaded cost of about **£78,000** · salary plus employer costs plus tools plus the recruitment, which you had not counted.

**Without the note**, you read £78,000, think *fair enough, still a saving*, and spend the next hour refining a case built on somebody else's number.

**With the note**, you have a much better problem. You said sixty. It says seventy-eight. That gap is not a rounding difference · one of you is wrong about something structural, and four minutes will tell you which. It turns out you were both wrong in the same direction, because neither of you counted the two days a month you currently spend managing the agency.

Your number was worse than the machine's, and that is not the point. **The point was having a number at all**, because a difference is visible in a way that a single figure never is, and there is no difference unless you wrote yours down first.

You do not need to be right. You need to have committed, so that being wrong is something you can see.

**Where it stops working:** if you genuinely have no view, a made-up number is noise rather than an anchor. The test is whether being wrong would surprise you. If it would, write it down. If it would not, skip the note and write what you expect the **second** answer to say instead.

### The two hours that stopped mattering

One more, and it is the other error these tools have quietly made worse.

Two hours into an approach that is not working. You know those two hours are gone and should play no part in what you do next. You feel the opposite, and the feeling usually wins, because stopping makes them wasted and carrying on means they might not have been.

Here is what changed. **When starting again was expensive, sunk cost was at least an honest calculation.** You were weighing two hours against a week of redoing it, and continuing was often correct. Now starting again costs four minutes.

The economics collapsed and the feeling did not move at all. Which means the pull you feel to continue is now attached to nothing, and it is the same pull, at the same strength, as when it was rational.

**The question that cuts it:** arriving fresh today, with no history, is this the approach I would choose?

If not, the two hours are gone either way. The only live question is whether you spend a third on top.

### Predict, then check · the habit that builds everything else

The most valuable thing in this chapter takes one line and almost nobody does it.

**Before you send something consequential, write down what you expect to happen.**

*I think she comes back with two objections, one about cost, one about timing, and asks for it by Thursday.*

One line, in a note, in a file you keep. Then when the reply arrives, look.

Do this twenty times and you'll have something almost nobody has: **a record of how well you actually read situations.** Not an impression. A record. And you'll find, as everybody does, that you are reliably good at predicting some things and reliably poor at others, and that the pattern is not the one you would have guessed.

### What the record looks like after a month

Six lines. That is the whole thing, and each one costs eight seconds.

> | Sent | I expect | What happened | |
> |---|---|---|---|
> | Budget paper to FD | Two objections · cost and timing. Wants it by Thursday. | Cost, yes. Timing never came up. Asked for it in two weeks. | **half** |
> | Process change to the team | Priya pushes back, the rest go along with it | Priya was fine. Marcus raised three things in writing that evening. | **wrong** |
> | Supplier note to ops director | He forwards it to procurement without comment | He did exactly that, within the hour | **right** |
> | Headcount case to the board | They ask what happens if we do nothing | They asked what it costs to delay six months | **half** |
> | Draft policy to legal | Comes back covered in changes | Came back with two, both about one clause | **wrong** |
> | Pricing to the FD | Argues the margin, not the volume | Argued the margin | **right** |

Two right, two half, two wrong · and the average is the least interesting number on the page.

**You read your seniors well and your peers badly.** Every line involving somebody more senior is right or half right. Both outright misses are lateral · Priya, who you were braced for, and legal, who you had catastrophised. You have spent years learning to predict the people who assess you and almost none on the people beside you.

**And you over-predict resistance.** Three of the six expected more pushback than arrived, which is expensive rather than charming: a person expecting a fight writes defensively, and defensive writing is longer, hedged and harder to say yes to.

Neither of those is a thing anybody could have told you. It is not in a review and your manager cannot see it. It is visible only because you wrote the guess down before you knew the answer, and the guess is worthless the moment you have read the reply.

One line before you send. Eight seconds, and it is the only feedback loop you have that isn't filtered through people being polite.

### What this actually buys you

Two errors, both made worse by the speed of these tools, and both cheap to defend against.

*What did I expect, before I looked?*
*Arriving fresh today, would I choose this approach?*

Then one line on what you think happens next, kept somewhere you will see it again.

Under two minutes, and between them they check the one participant in this arrangement that nothing else does. You can brief perfectly, check exhaustively and score honestly, and still be confidently wrong in a way no amount of checking will reach, because every one of those checks runs on a question you chose.

---

> ### ▪ DO THIS
>
> **Two questions and a prediction, before the next thing that matters.**
>
> **1. Write your expectation before you look.** One line, thirty seconds. Now you have your own number to compare against instead of inheriting theirs.
>
> **2. When the first answer was substantial, open a fresh window for the second opinion.** Not because the tool remembers. Because you do.
>
> **3. Ask the fresh-start test.** *If I arrived today with no history, would I choose this?* If no, the hours already spent are gone either way.
>
> **4. Then one line on what you expect to happen.** *She comes back with two objections, cost and timing, and asks for it by Thursday.* Keep it where you will see it again.
>
> **5. When the reply lands, open the note.** Write *right* or *wrong* and one word on why. Ten seconds, and it is the step that makes the other four worth doing.
>
> **Under two minutes. Steps 4 and 5 take thirty seconds between them.**
>
> **When you have no time for any of it:** do step 4 only. It is the one that compounds.
>
> **You'll know it worked when** your first prediction turns out wrong in a way you can name. Because that is a thing you learned about yourself that nobody could have told you.

---

Two questions and a prediction, and every one of them needs you to remember to ask.

That is the whole problem with this chapter and with the ten before it. Nothing anywhere is going to remind you.


## 12 · Nothing Will Remind You to Think

Somebody went and asked, on real people doing real work, and the finding is worse than you would guess.

In 2025, researchers at Microsoft Research and Carnegie Mellon surveyed 319 knowledge workers who used these tools at least weekly, and collected 936 first-hand examples of them doing it. They were looking for when people actually think critically about what comes back, and what makes them stop.

The headline result is one sentence:

> *"higher confidence in GenAI is associated with less critical thinking, while higher self-confidence is associated with more critical thinking."*

Read that twice. What they found is an association, in self-reported data, and I want to be exact about that before I build anything on it. They did not watch confidence cause the drop. They asked people, and the two moved together.

Here is what I think it describes, and you should treat this next part as my reading rather than their finding.

The better the tool gets, the more you trust it. The more you trust it, the less you check. The less you check, the less practised you are at checking, so the next time you are even less equipped to notice, and the tool by then is better again.

That is a trap with no floor in it. The survey is one turn of that loop, caught in a single moment, on 319 people. Whether it runs the way I have just described is the thing you can test on yourself, and this chapter is mostly about how.

Nothing in that loop contains a moment where something stops you. There's no alert. No amber light. **The output does not look different on the day it is wrong.**

That is the argument of this chapter, and it's the one I'd keep if you made me throw away every other technique in this book. Every other chapter is a method. This one is a habit, and habits are the things that decay when nobody is watching.

### The two ways to get this wrong, and both of them are failures

Here is where most writing on this becomes useless, because it only warns you about one direction.

The obvious failure is trusting it too much. You accept something that was wrong, it goes out with your name on it, and you find out later in a room you did not want to be in.

The failure nobody mentions is the opposite one. **You reject something that was right.** You spend an hour re-doing work that was already correct, you disregard an objection that would have saved you, or you decide the whole category is unreliable and go back to doing everything by hand while somebody two desks away produces four times as much.

Human factors research named both of these nearly thirty years ago, well before any of this. Parasuraman and Riley, in 1997, called them **misuse** and **disuse**: over-reliance on automation on one side, neglect of automation that would have helped on the other. The modern literature on working alongside these systems calls the same pair over-reliance and under-reliance, and it's consistent that both damage the quality of your decisions.

So this isn't a chapter telling you to be sceptical. Scepticism applied uniformly is just a slower way of being wrong, and it burns the hours the tools gave you back.

**It's a balance, and the balance is the skill.** Accept what's right. Test what you're unsure of. And spend the time you saved on the thing neither of you has thought of yet.

### Why the balance tips the wrong way on its own

Left alone, this does not stay balanced. It drifts one way, and it's worth understanding why, because the reason is not laziness.

A well-formed answer removes the felt need to check it.

Look at what arrives when you ask one of these things a real question. It's structured. It's calm. It anticipates the obvious objection and deals with it in the third paragraph. It uses the vocabulary of your industry correctly. Every single one of those properties is a signal your brain has spent a lifetime reading as *this has been thought about by somebody competent*.

None of those properties has any connection to whether it's true.

That is the whole mechanism, and it's the same one as chapter three, arriving at a different point in the process. In chapter three it was a feeling of relief that moved your hand. Here it's the shape of the answer doing the same job. Both of them produce the identical outcome, which is that you move on.

Do that once and nothing happens. Do it for a year and something does.

### The three moves

Here is the discipline, and it is three moves in a fixed order. Not the three questions from chapter one · those interrogate an answer you have already got, and these decide what to do with it. It takes about ninety seconds and I would rather you did it badly than skipped it.

**1. What here can I accept?**

Start with acceptance, deliberately, because starting with doubt makes this exhausting and you'll stop inside a fortnight.

Most of what comes back is fine. The structure is fine. The gathering is fine. The arithmetic is usually fine. Accept it, out loud, and notice that you have made a decision rather than drifted into one.

The test for this is not *does it feel right*. It's **would I have been able to produce this myself, and does it match what I already know to be true?** Where the answer is yes to both, accept and move on. That is not laziness, it's the appropriate use of a thing that is genuinely better than you at that particular stage.

**2. What am I unsure about, and what's the cheapest test?**

Now find the parts you can't accept, and be specific. Not *I don't fully trust this*. Which sentence.

Then match the test to the stake, which is chapter five's whole argument arriving here in a different form:

- A figure you'd repeat in a meeting. Ask where it came from and go and look at the source. Two minutes.
- A conclusion you'd act on. Make it argue the opposite, which is chapter seven. Twenty seconds.
- Something you'd send to somebody who matters. Put it through a second system, which is chapter nine. Ninety seconds.
- A claim you can't check at all. Say so, in the document, in your own words. That sentence costs you nothing and it's the one that protects you.

Notice that none of these is *are you sure*. Chapter seven explains why that question is worthless, and it's the question almost everybody asks.

**3. What has nobody thought of?**

This is the one that's actually hard, and it's the one worth your career.

The machine is genuinely good at half of it. Ask it directly what you have failed to consider and it will produce things you had not thought of, on any subject, reliably. Most people never ask, which is astonishing given how cheap the question is.

The other half is the one chapter two described · the absences nobody ever wrote down, which is precisely why nothing that reads can find them.

So run it in both directions, and the order matters. Ask the machine what is missing from the material. **Then** ask yourself what is missing from the situation. Doing it that way round means you are not competing with it, you are picking up where it stops, and you will be surprised how often its list makes yours obvious.

### The three moves on one real thing

Ninety seconds is easy to say. Here is the whole of it on a single ordinary output, so you can see how little of it is work.

You asked for an analysis of why complaints rose last month. Back comes two pages: volumes by category, a rise concentrated in category three, three candidate causes, and a recommendation to review the escalation process.

**What can I accept?** The volumes, because they came out of the system and you can see the query. The arithmetic, because you spot-checked two rows and they hold. The observation that category three carries the rise, because you can see that yourself in ten seconds. That is most of the document, accepted deliberately, in about twenty seconds · and notice that you have now *decided* to accept it rather than drifted past it, which is the entire difference this chapter is about.

**What am I unsure of, and what is the cheapest test?** One sentence, and you can name it: *the rise correlates with the new booking flow going live on the 8th.* You are unsure because correlate is doing a lot of work in that sentence and you know the flow went live on the 8th because you were in the meeting.

Cheapest test, matched to the stake: this is going to your director and it will drive a decision about the booking flow, so it earns more than one pass. *Show me the daily complaint counts for category three for six weeks either side of the 8th, and tell me what else changed in that window.*

What comes back is that a second thing changed on the 11th. A supplier switch nobody had mentioned. And the daily counts move on the 11th, not the 8th.

**What has nobody thought of?** Run it both directions, which takes forty seconds. Ask the machine: *what would explain this rise that is not in the material I gave you?* It produces four things, one of which is seasonality you had not considered.

Then ask yourself, which is the half only you can do. And the answer is sitting there: **complaints in category three are logged by the same two people who were on annual leave for the first week of the month.** So the first week is not low. The first week is unrecorded, and the "rise" is partly a return to normal logging.

**Ninety seconds, and the document changed twice.** Once because the date was wrong, and once because the baseline was wrong · and the second one would have survived every check in this book except the one only you can run.

### What to interrogate, when you don't know where to start

Richard Paul and Linda Elder spent a career on this and produced a framework used in universities everywhere. It has eight elements of reasoning and nine standards to judge them against, which is more than anybody remembers under pressure.

So take four of them, which is what actually survives contact with a Tuesday.

**The purpose.** What is this piece of work for? Not the task. The decision it feeds. Half of all bad output is a correct answer to the wrong purpose.

**The question.** Is the question it answered the question you needed answering? These systems are relentlessly obliging and will answer a nearby question beautifully rather than tell you yours was ill-formed.

**The assumptions.** What has been taken as given? Every answer rests on something unstated, and the unstated part is where the error lives. Ask for it explicitly: *list the assumptions this depends on.* It will.

**The consequences.** If this is wrong, who finds out, and when? That's chapter fourteen's question and it belongs here too, because the answer determines how much of this discipline the work deserves.

Four questions. Purpose, question, assumptions, consequences. You can hold that in your head in a lift.

### It has to be a discipline, because nothing else will prompt you

Everything else in this book has a trigger. A brief gets written when work arrives. A score gets asked for when something looks finished. A second system gets opened when a decision matters.

This one has no trigger. Nothing in the design of any of these tools will ever say *you have accepted eleven things in a row without checking one of them*. The interface is built to be helpful, and a prompt to doubt it would be a strange feature for anybody to build.

Which leaves you. Not your judgement in the abstract, but a specific habit you decide to run.

Mine is the ninety seconds above, and I run it on anything I would put my name to. That is a low bar and it's deliberately low, because a discipline you actually keep beats a better one you abandon.

Yours might be different. What it cannot be is nothing, and *I'll notice if something looks wrong* is nothing, because the entire finding of that survey is that the better this gets, the less anything will look wrong.

---

> ### ▪ DO THIS
>
> **Take the last thing you accepted without checking. Ten minutes.**
>
> **1 · Find it.** The most recent piece of output you used, forwarded or acted on without going back over it. There will be one from this week.
>
> **2 · Run the three moves in order.** What can I accept. What am I unsure of and what's the cheapest test. What has nobody thought of.
>
> **3 · Then ask it directly:** *List the assumptions this answer depends on, and mark any you cannot support.*
>
> **4 · Go and check exactly the ones it marked.** Not all of them. The marked ones.
>
> **Ten minutes the first time, ninety seconds after that.**
>
> **If you find nothing wrong:** that's the good outcome and it's not a wasted ten minutes. You now have the right to put your name on it, which you did not have before you looked.
>
> **You'll know it worked when** you catch yourself doing move one deliberately, on something you would previously have accepted without noticing you had.

Ten chapters told you how to get more out of these tools. This one and the one before it are about the person doing it, because that is the part with no upgrade path and no support contract, and it's the only part your employer is actually buying.

Which covers a single exchange, thoroughly, from both ends. And almost nothing you are actually paid for is a single exchange.


## 13 · Decide It Once, Then Keep It Honest

Real work is six steps where the fourth depends on what the second turned up.

Almost nobody has been shown how to run that with a machine doing the steps, so people do the thing that feels responsible. They plan the whole job at the start and then follow the plan through material that has been telling them since step two that it was the wrong plan.

Everything in this book so far has been a single exchange. Ask, check, score, send. This chapter is about the longer jobs, which is most of what you're actually paid for.

### The plan you write first is the worst plan you will ever have

Not because planning is bad. Because of when it happens.

At the moment you write a plan you've less information about the job than at any later point. Every step adds something. So the plan is authored at peak ignorance and then defended, by you, against everything you learn afterwards, because changing it feels like conceding the first version was wrong.

The alternative isn't to abandon planning; it is to move the planning **between the steps instead of before them.**

Do the first step. Read what came back, properly. Then ask: *given what I now know, what is the most useful next step?*

That question, asked between every step, is the whole method. Ten seconds each time, and it's the difference between six steps that compound and six steps chosen by somebody who had not started yet.

I keep arriving at this shape. I have built research systems three times, for different purposes, and ended up with the same architecture each time. Run one step, read the result, choose the next from what is now known. Treat that as my habit rather than as proof. The argument above stands without it.

### Decide the finish line before you start

**I stop when I get bored.**

So does everybody I have watched. Not when the question is answered. When the afternoon runs out, or the reading turns repetitive, or something louder arrives. And because it feels like a natural ending rather than a decision, almost nobody counts it as one, or notices they made it somewhere different last week on a similar job.

You have met this shape three times already, and it is worth naming now rather than letting it keep arriving in disguise. Chapter five decided the depth before asking. Chapter eight wrote the bar down before seeing the score. Chapter eleven noted the expectation before looking at the answer. **Every one of them is the same move: commit while you are still able to, because after you have seen the thing you cannot.**

This is the fourth and the most expensive, because it is the one measured in hours rather than minutes. Write the finish line down before the first step. Written down, not intended.

The rule most people reach for is **three independent sources**, and it has a hole in it. Three outlets carrying the same wire story is one source wearing three coats. Independent means a different origin, not a different website, and if you can't say where each one got it, you've one.

The same hole sinks the other obvious rule, *the number stops moving*, because a number stops moving precisely when nothing new is being checked.

So the condition has to be both halves at once: **two consecutive passes turn up nothing I did not already have, and I can say where each source got it.** Written at the top of the page, before the first step.

Then, when you feel the pull to stop, you have something to check the feeling against. Sometimes the honest answer is *not yet*. More often the condition was met a while ago and you were continuing out of anxiety, which is an hour you have just been handed back.

### Look before you build

Before planning anything: **have I already solved part of this?**

Somewhere on your machine is an instruction you wrote three months ago that does a chunk of what you're about to start from scratch. A brief that worked. A structure built for a different client that transfers exactly. A checklist you made after something went wrong.

You won't remember it, because you filed it where you file things, and that place is a graveyard.

Two minutes at the start of anything. Look before you build. And the reason this fails isn't laziness — nothing was ever saved in a form that could be found again.

### The judgement, made once

I have built sixteen applications that I use, not demos. I don't write code professionally and never have.

That isn't the point, and I'm not telling you to build sixteen of anything. The point is available to you this week without writing a line of anything.

**Every one of those tools is a judgement I made once and then stopped making.**

How to tell whether an article is good enough. What makes an event worth the flight. When research has gone far enough. Each was a decision I used to make freshly every time, slightly differently depending on how tired I was and what had happened that morning.

Now each is written down in a form that runs.

The speed is real and it isn't the interesting part. **The consistency is.** The standard I applied on a bad Thursday in February is the standard applied now, because I'm not the one applying it any more. I am the one who decided it.

That has almost nothing to do with software.

### What yours looks like

Not an application. An instruction, with your actual standards in it, in the words you would use to a competent colleague doing it for you.

Here is a complete one. It is dull on purpose, because the dull recurring judgements are where this pays.

> *You are a procurement manager with fifteen years in a mid-sized services business. You are looking at a supplier quote to decide one thing: is this worth an hour of my time in a meeting, or is it a no?*
>
> *Check these five, in this order.*
>
> *1. Does the quote answer what we actually asked for, or a nearby question they preferred? Quote the line that shows it.*
> *2. Is the price broken down enough that I can see what is being assumed? A single number is a flag, not a price.*
> *3. What is excluded? List anything a reasonable buyer would expect to be included and is not.*
> *4. What are they committing to and what are they merely describing? Separate the two.*
> *5. What would have to be true for this to go wrong at month four?*
>
> *Then give me: a verdict of MEET, ASK FIRST, or NO. If ASK FIRST, the single question that would settle it, in under twenty words. If NO, the one line I can send them.*
>
> *Stop when you have all five. Do not go looking for a sixth thing.*
>
> *Do not summarise the quote back to me. Do not give me options without a verdict.*

Note the second-to-last line. **The stopping rule lives inside the instruction**, not in your intentions, because that's the only place it survives a busy week.

### And now the part that makes it a job rather than a task

That instruction is one step. Watch what happens when you actually use it, because this is the shape the whole chapter is about.

It comes back **ASK FIRST**, with one question: *does the implementation fee cover the data migration or not?*

Here is the moment. You don't go to step two of a plan, because you never wrote one. You ask: *given what I now know, what is the most useful next step?*

And the answer has changed. Before you ran it, the obvious next step was a reference check. Now it isn't, because the whole decision has collapsed onto one ambiguity, and a reference won't resolve it. **The finding chose the next move.** You send the question.

They come back and the fee doesn't cover migration. Ask again. Now the useful step isn't a meeting either; it's a rough number for what migration costs, because if that number is large the quote you were assessing was never the real price and the comparison you were about to run was wrong.

Three steps. None of them planned in advance. Each chosen from what the one before it turned up, and the second and third would both have been wrong if you had decided them at the start.

That is the method, and it costs one question asked out loud between each step.

The instruction takes roughly twenty minutes to write. Use it three times and you'll notice something missing, and you'll add it. **That is the moment it stops being a note and becomes an asset**, because it now holds something you learned rather than only what you already knew.

Give it a file of its own, named for the judgement in the words you would use to a colleague, in one place with the others. That is what *a form that could be found again* means, and it's the whole of what the two-minute look needs to work.

Do that a few times over a few months and you have several of them. They take twenty minutes to draft and a good while of use to become worth anything, which is exactly why almost nobody has them.

### Two things that are not your call

I am not going to hand you this without the part that gets people into trouble.

**What your contract says about work product.** In most places, depending on your contract, things you create in the course of employment belong to your employer. That is usually fine and occasionally matters, and you want to know which before you build something you think of as yours.

**What your employer's policy says** about putting internal standards, client material or process detail into an AI tool. Many organisations now have one. Some have one nobody has read.

Do this in the open. Tell your manager you're building it. It is worth more to you visible anyway — a private asset makes you faster, and a visible one makes you the person who improved how the team works, and only one of those gets discussed in a pay conversation.

### Where this gets uncomfortable

The obvious objection is the right one.

**If you write the judgement down, is it still yours?**

You saw this in chapter two. The moment a rule leaves somebody's head and gets documented, it becomes rule work, and rule work is the work that moves away from people.

The distinction is precise. **What you write down is the standard. What you keep is deciding it.**

Anybody can run your checklist. Almost nobody can build it, and the person best placed to notice it has stopped being right is the one who set it. The judgement in the file is the judgement you made in March. The valuable thing is being the person who notices in September that March was wrong.

Which is why the title of this chapter isn't *do it once and never again*. The asset has to be maintained or it quietly becomes a liability, and **a checklist nobody has questioned in two years is somebody's outdated opinion being applied with great consistency.**

That is how good judgement actually decays, and it's the honest cost of everything here.

### Watch it happen to the instruction I just gave you

Take the supplier-quote instruction from a few pages back and run it forward eighteen months, because this is the part nobody warns you about and it is the reason most of these things quietly stop being worth having.

It works. That is the problem.

It works so well that after the first three months you stop reading its reasoning and start reading only the verdict. MEET, ASK FIRST, NO. By month six you are forwarding the ASK FIRST question to the supplier without opening the rest. By month twelve somebody else on the team is using it, because you gave it away, and they never saw the reasoning at all · they inherited a verdict machine.

Now look at check number two. *Is the price broken down enough that I can see what is being assumed? A single number is a flag, not a price.*

That was right when you wrote it. Then your industry changed the way it quotes. A fixed all-in price became the normal, competitive, customer-friendly way to sell this thing, and the suppliers still itemising are the ones who intend to add to it later.

**Your rule has inverted.** It now flags the good quotes and waves through the bad ones, and it does this with total consistency, every time, in your name, and it has never once told you it was struggling. A person applying that rule would have said *this feels wrong lately.* A file does not have that feeling.

Then the second failure, which is worse and quieter. Check five asks *what would have to be true for this to go wrong at month four?* Nine months in, a supplier failed at month four in a way nobody had seen before · they were acquired, and the team that had written the quote left. You had a bad quarter because of it, you learned something real, and **you did not put it in the file.** You put it in your head, where it will stay until you leave.

So the file is now two things at once. It is a rule that has quietly gone backwards, and it is missing the single most expensive thing you learned in the period it covers.

**Here is the maintenance, and it is smaller than the problem.** Once a quarter, open the file and answer two questions.

*Which of these checks has fired in a way I disagreed with?* Any check you have overridden more than twice is not a check, it is a formality, and either the rule is wrong or you are wrong. Both are worth knowing and you cannot find out without looking.

*What did I learn since last time that is not in here?* Not everything. The one thing that cost you something.

Twenty minutes a quarter. That is the whole maintenance schedule, and it is the difference between an asset and a fossil.

**And the thing that makes it work is a date.** Put the date you last reviewed it at the top of the file, where you cannot miss it, because a rule with a date on it is a rule somebody can question. Anybody who opens it can see it is fourteen months old and treat it accordingly. A rule with no date on it reads as permanent, and nothing in this book is permanent.

That is also the honest answer to the objection I raised a moment ago. What you keep is not the standard · the standard is in the file and anybody can run it. **What you keep is being the person who notices in September that March was wrong**, and the quarterly twenty minutes is what that noticing actually looks like when it is a habit rather than a virtue.

### The common mistake

Almost everybody writes the instruction and never writes the stop.

The instruction is the interesting part, so it gets the attention. The stopping rule feels like a detail you can add later, and later does not come. What you are left with is a tool that runs until whoever is watching decides there has been enough, which is the exact thing you built it to eliminate · your own boredom, wearing a more professional coat.

If you write only one line of it today, write the line that says when to stop.

### This week

**One thing.** The judgement you make most often.

Write it as an instruction, with your real standards, in about twenty minutes. Put its stopping rule inside the file before you use it once. Check the two things above first. Use it three times, asking between each one what the last result changed. Add what you notice is missing.

At the end you'll apply one standard the same way twice in a row, which you almost certainly didn't do last month.

That is a smaller claim than I would like to make and it's the one I can support.

And it puts you somewhere specific. Telling your manager you're building it isn't the same as having it checked, and you're now producing consequential work through a process nobody has tried to break.

So give it to one person whose job it's to disagree with you, and ask them to break it. That is the last cheap thing available to you before the expensive question arrives, which is what could go wrong. And almost nobody in your building has said it out loud.


## 14 · Ask What Could Go Wrong

What you type into an AI chatbot isn't privileged. Not now, and on the court's reasoning, not ever.

In February 2026, a judge in New York answered a question he believed no court had been asked before.

A man had been indicted. Securities fraud, wire fraud, and more. After he received the grand jury subpoena, and after it was clear he was the target of the investigation, he did something that will feel completely ordinary to you.

He opened an AI chatbot and started thinking out loud.

He worked through what the government might charge. He drafted what he might argue. He talked himself through his own situation, the way you would with a colleague you trusted, except there was no colleague. His own lawyer later confirmed nobody had told him to do it. In the court's phrase, it was done *"without any suggestion from counsel that he do so."*

Then the FBI executed a search warrant at his home and seized, among everything else, **31 documents containing those conversations.**

His lawyers argued they were privileged. Confidential. The sort of thinking a person does preparing a legal defence, protected the way notes to your own lawyer are protected.

Judge Jed Rakoff wrote that this appeared to be *"a question of first impression nationwide."* Nobody had ever asked a court whether what you type into an AI chatbot is privileged.

His answer, in five words: **"the answer is no."**

### Why it was no, and why it applies to you

You are not under federal indictment. Stay with me anyway, because the reasoning is not about him.

The court gave two grounds, and the first disposed of the case on its own.

**It is not a lawyer.** Obvious once said, and yet nobody thinks about it while typing. The court went past the technicality, and this is the sentence to keep. Recognised privileges require *"a trusting human relationship"* with *"a licensed professional who owes fiduciary duties and is subject to discipline."* And then:

> *"No such relationship exists, or could exist, between an AI user and a platform such as Claude."*

Not *does not yet.* **Could not.**

**And it was not confidential.** The court read the privacy policy the user had agreed to. It provides that the company collects data on inputs and outputs, uses that data to train the system, and reserves the right to disclose it to *"a host of 'third parties,' including 'governmental regulatory authorities.'"* Even without a subpoena compelling it.

Then the line that reaches past this one case, the court quoting another court. Users *"do not have substantial privacy interests in their conversations with [a publicly accessible AI platform] which users voluntarily disclosed to the platform and which the platform retains in the normal course."*

**Voluntarily disclosed.** That's what your typing is, legally. Not private thinking that happens to occur in a text box. Disclosure, to a third party, retained in the normal course of their business.

I should be accurate about the limits, because this chapter is about not overclaiming. One district court, on privilege, in a criminal matter. It binds nobody elsewhere and doesn't say chat logs are public. But the court believed no one had answered the question before, and the reasoning was not narrow.

The same opinion notes, from published research, that more than half of American households now use these tools in some form, and that one platform alone is used by **more than 800 million people a week.**

Almost none of them know what that judge decided.

### The second case

In January 2026, in the same district, another judge affirmed an order in a copyright case against an AI company.

The order requires production of **twenty million conversations.**

Not twenty million belonging to the parties. Twenty million belonging to ordinary people who had nothing to do with the lawsuit, had never heard of it, and were never asked.

Be precise about this one, because the version circulating is more alarming than the truth. The sample is **de-identified**, and a **protective order limits who can see it.** Your account name isn't attached, and a stranger can't go and read it. Though anything identifying that you typed into the conversation is still sitting inside it.

That still leaves the fact. Twenty million private conversations became discoverable material in litigation between other people, and every decision about them was made in rooms none of those twenty million were in.

The exposure isn't that somebody is reading your messages. It is that **what happens to them is decided entirely by people who are not you, in proceedings you will never hear about.**

### The ten questions

None of these is technical. There's the point. Every one can be asked by somebody who's never written a line of code, and asking three of them will change how a room sees you.

**1. Where does this go when I press enter?** Which company, which country, which servers. Most people using a tool daily cannot answer this about the tool.

**2. Is what I type used to train it?** Usually a setting. Usually the default isn't the one you would choose. Almost nobody has looked.

**3. How long is it kept, and who decides?** Not what the marketing page implies. What the terms say, and whether the company can change them next quarter without telling you.

**4. Under what circumstances would they hand it over?** The privacy policy in that New York case answered this in advance, in writing, and everybody had agreed to it.

**5. Whose account is it under?** A personal account and a company contract are different legal objects with different terms. If your team is doing work in personal accounts, your organisation has obligations it doesn't know about.

**6. What is the most sensitive thing already typed into it?** Not hypothetically. Actually, by your team, last month. Almost no manager can answer this, and it is not a policy failure. Nobody built a way to know.

**7. If we had to reconstruct how this was made, could we?** Which question was asked, which system answered, what came back. If a regulator, an insurer or a client asks how a piece of work was produced, *someone used AI and then edited it* isn't an answer.

**8. What happens if this tool disappears in six months?** Companies retire products. If your process only works because one specific thing exists, that's a dependency, not a workflow.

**9. What is the worst single output that could reach a customer?** Not the likeliest. The worst. Then: what stands between that output and the customer, and is it a system, or is it somebody remembering?

That question has a failure mode worth guarding against, which is that you will answer it from whichever direction you were already worried about. Ask it seven ways instead; it takes a minute and it will find something the single question doesn't.

**Emotional**. Could this upset, offend or embarrass somebody.
**Legal**. Could this create an obligation, a breach, or something quotable in a dispute.
**Financial**. Could this cost money directly, or commit us to something.
**Relationship**. Could this damage a connection we depend on.
**Timeline**. Could this be late in a way that matters, or create a deadline we can't meet.
**Competitive**. Does this hand anything useful to somebody we compete with.
**Information**. Does this reveal something we did not intend to reveal.

Seven categories, most of which will be empty on any given piece of work. The value is entirely in the ones you wouldn't have looked at, and for most people that's the third and the seventh.

**10. Who is accountable when it is wrong?** If the answer is *the AI got it wrong*, there's no answer, and everybody in the room already knows it.

### Why asking these gets you into rooms

I want to be direct about the career mechanics, because that's what this book is for.

Senior people are worried about this and most can't articulate why. They sense that something is being adopted fast, that the risk is real, and that everyone briefing them is either selling something or explaining the technology instead of the exposure.

Walk into that meeting and ask question six. *What is the most sensitive thing our team has already typed into one of these?*

You have not claimed expertise. You haven't needed any. You asked the question the room was circling, and you will be the person they think of next time, because you did the thing nobody else did, which was to come at it from the direction of what goes wrong.

There's the whole move. Available to you today, and it costs nothing.

### The part where I stop making it sound solvable

The instinct after reading this is to go and find the safe tool. The compliant one, with the right badges.

Wrong move, and it's the most common one.

Every one of these tools is a product, made by a company that will change, be sued, be acquired, or update its terms. Picking the current best one is a decision with a shelf life, and it puts your protection in the hands of somebody whose incentives aren't yours.

The better move is the boring one. **Change what you send.** Not whether you use it. What it receives.

If the sensitive part never leaves, you don't have to trust anybody's terms of service about the sensitive part. That is the only protection in this chapter that doesn't depend on a company continuing to behave well.

And be honest about what it buys, because the overclaiming version of this argument is its own kind of failure. It reduces what leaves your control. It doesn't erase your obligations, doesn't make you compliant, and doesn't replace somebody reading the output before it goes out. **Reducing what leaves your control is where every privacy framework starts**, and it is a thing you can explain to a nervous client in one sentence.

There is a working rule inside it, smaller than a policy. If you would not be comfortable with the sentence you are about to type appearing in a legal filing with your name on it, do not type it. Rewrite it so the identifying part isn't there. Four seconds, and that is the whole discipline.

Here is what four seconds looks like.

You are about to type this:

> *Draft a firm but fair note to Sarah Chen at Meridian about the overdue invoice, reference 7741-B, for $240,000, and mention that her team missed the same deadline in March.*

You type this instead:

> *Draft a firm but fair note to a long-standing client contact about an overdue six-figure invoice, and mention that their team missed the same deadline earlier this year.*

Read both. The second one produces **a letter you can use without changing a word of the argument.** None of the useful instruction was in the name, the reference number, or the amount. You put those back yourself, in about ten seconds, in the document where they belong.

That is the whole idea and it's smaller than people expect. The identifying details are almost never the part doing the work. They feel essential because they're what the task is *about*, but the tool isn't writing about Sarah Chen. It is writing a firm but fair note about an overdue invoice, and it will do that just as well without ever learning who she is.

Try it once on something real and you will notice the same thing everybody notices: the output doesn't get worse.

### Why a policy will not save you

One more thing worth understanding, because it explains why every attempt to fix this with a memo fails.

Most organisations respond to all of the above by writing a rule. *Do not put client information into AI tools.* It goes in the handbook. Everybody nods.

Then people use the tools anyway, on their phones, on personal accounts, with no controls at all, because the rule made the tools useless for the work that actually matters and the work still has to get done.

**That is not a failure of policy. It is a failure of design.**

A rule that forbids the only version of the task worth doing does not get followed. It gets routed around, invisibly, by people who aren't being defiant. They are being practical, and now the same exposure exists with none of the visibility.

The fix is not a stricter rule; it is changing what gets sent, so the rule becomes unnecessary. When the sensitive part never leaves in the first place, there's nothing to forbid, and the working instruction flips from *do not paste the important stuff* to *paste it, it's handled.*

That flip is what brings the real work into scope. And the real work is the only work that made any of this worth adopting.

### Nobody decided anything

There is one failure mode in all of this that I want you to see clearly, because it is the one that actually happens and it looks nothing like negligence.

I know of a meeting-transcription tool, built by a team who had already written redaction layers into their other products, that went out without one.

Meeting transcripts are among the most sensitive text any organisation produces. Personnel discussions, commercial terms, things said out loud that nobody would ever write down.

Nobody argued for leaving the protection out. Nobody raised it and was overruled. There is no meeting where a decision was taken.

The thing got built for what it does. The privacy layer was somebody's job later. **And later does not arrive on its own.**

Nobody decided to ship it without protection. **Nobody decided anything.**

Which is exactly why the ten questions have to be asked **out loud, by a person.** Every one of them is a decision that will otherwise be made by default, and defaults are not decisions. They are just what happened while everybody was busy.

### The one to ask tomorrow

If you take one thing from this chapter, take question six, and ask it about yourself before you ask anybody else.

**What's the most sensitive thing you have personally typed into one of these tools?**

You will remember something. Everybody does. A client name, a salary figure, a contract clause, a paragraph about a colleague that you would never have put in an email.

It is still there. It was, in the words of a federal judge, voluntarily disclosed, and retained in the normal course.

Nothing bad has happened. Which isn't the same as nothing being at risk, and you now know the difference, which is more than almost anyone you work with knows.

Which leaves you somewhere strange; you are doing careful work now. Checked properly, scored honestly, produced safely.

All of which protects you.

---

> ### ▪ DO THIS
>
> **Ask yourself the question you are about to ask the room: what is the most sensitive thing already typed into one of these?**
>
> **1. What's the most sensitive thing you've personally typed into one of these tools?** Sit with it rather than moving on. You'll remember something.
>
> **2. Go and look at two settings** on the tool you use most: whether your inputs train it, and how long they're kept. Two minutes.
>
> **3. Rewrite one thing before you send it.** The name becomes *a long-standing client*. The number becomes *a six-figure invoice*. Send only that version, and compare what comes back against what you already had.
>
> **4. Take one question into your next meeting.** *What's the most sensitive thing our team has already typed into one of these?* Almost no manager can answer it.
>
> **Steps 1 to 3 take ten minutes. Step 4 goes into your next team meeting.**
>
> **When the answer to step 1 frightens you:** it has been disclosed. You may still be able to delete the record. Go and look, in the same settings as step 2. Then change what you type next week.
>
> **You will know it worked when** the outputs from step 3 come back the same, and you realise the identifying details were never doing the work.

---

None of which protects the company you work for, because while you were asking where your own typing goes, a great deal of law was quietly written about what your employer does with everybody else's.


## 15 · Your Responsibility With AI

*Eleven questions that will make you seem wise, or save your company*

The most available opportunity in this book is a set of questions that almost nobody in your organisation can currently ask, about a rule that's already in force.

On 2 August 2026, a set of obligations in European law began applying to AI systems used to make decisions about people. Hiring. Credit. Education. Access to essential services.

They apply to organisations **using** those systems, not only to the companies that built them.

I would like you to go and ask three people in your business whether that affects you. My prediction is that all three will say it's somebody else's department, and that not one of them will be able to name the department.

That gap is the subject of this chapter. Not because you're going to become a lawyer. Because the questions that matter here aren't legal questions. They are management questions with legal consequences, and the people who can ask them are almost nowhere.

I am not a lawyer, none of this is advice about your situation, and every question below exists to send you to somebody who can answer it properly rather than to replace them. That isn't a disclaimer bolted on the front. It is the actual claim of the chapter. **The value is in asking, not in knowing.**

### The shift nobody has explained to you

For most of the last decade, the assumption was that AI regulation would land on the technology companies. They build it, they're enormous, they can afford it, and they're where the headlines point.

That assumption is now wrong, and being wrong about it is expensive.

The law has moved towards the **deployer**. The organisation that takes a system somebody else built and points it at a real decision about a real person; that is your employer. It is possibly your team. In one European framework the obligations on a deployer sit in their own article, separate from the builder's, and they include running an impact assessment on people's fundamental rights before you switch it on.

Nobody is going to send your company a letter about this.

**Read that clause again slowly.** Your organisation can buy a tool, use it exactly as instructed, and be the party carrying the obligation. The vendor's compliance isn't your compliance. It never was, and now it's written down.

### The four things happening at once

I am going to give you the shape rather than the detail, because the detail changes and the shape doesn't.

**Europe built a single law and phased it in.** Prohibited uses and a staff-competence duty first. Then general-purpose models. Then the high-risk categories, which is where most ordinary businesses actually live. Penalties at the top run to **thirty-five million euro or seven per cent of worldwide annual turnover**, whichever is greater.

Seven per cent of turnover. Not profit. Turnover.

**The United States built no single law and is producing dozens.** The National Conference of State Legislatures, which counts these, records that in the 2025 session thirty-eight states adopted around a hundred AI measures out of roughly twelve hundred bills introduced. By March of 2026, on the same count, forty-five states had introduced **1,561 AI bills**. Already past the total for all of 2024, before the year was half done.

They don't agree with each other. One state gives you an affirmative defence if you've adopted a recognised risk framework. Another gives individuals a **private right of action**, which means you aren't waiting for a regulator, you're waiting for a plaintiff.

**Australia looked at a dedicated AI Act and decided against it.** In December 2025 the government formally abandoned mandatory AI guardrails in favour of a technology-neutral approach: clarify the existing law, add guidance, fund a safety institute.

Everyone read that as the light-touch option. It is the opposite. A dedicated AI Act gives you a checklist. Technology-neutral means **every existing law applies to your AI the same way it applies to everything else**. Consumer protection, misleading conduct, discrimination, privacy, directors' duties. And there's no single document to read.

**The United Kingdom has no AI Act either**, and works through five principles applied by existing regulators within their own remits. No single AI regulator to call.

### And the liability question resolved in the least comfortable way

There was a proposed European directive dealing specifically with AI liability. It was **withdrawn** in early 2025.

That sounds like good news. It is not.

What happened instead is that the general product liability regime was revised to bring software, AI systems and digital services inside the definition of a **product**. Which means strict liability, disclosure obligations, and; this is the part to hold on to. **Rebuttable presumptions of defectiveness and causation for complex AI products.**

In plain English: for a system nobody can fully explain, the law has started shifting the burden of proof towards the people who deployed it.

The specialist law that would have been complicated to comply with was dropped. The general law that's much harder to argue with swallowed AI instead.

### Where I think this ends up

Here is a prediction. I'm marking it as mine so you can hold me to it.

There's a job that is going to exist in a few years and doesn't have a name yet. When a business uses AI, is that business legally responsible for what its AI does under its own name, or is nobody responsible? Today the honest answer is that it depends who you ask, which is another way of saying it hasn't been decided.

We have solved this before, in industries where the consequences arrived earlier than they did here.

In financial services, a named individual carries personal accountability for the way a firm conducts itself. On licensed premises, a named person is responsible for what happens inside the building. Not the company in the abstract. A person, named on a document, who can be asked.

I think that arrives for AI, and sooner than the people running businesses expect. Somebody in each company becomes the nominated person for whether AI is being used responsibly inside it, and they will have signed something saying so.

Companies will hate it at first. They should want it anyway, because the alternative is exactly where we are standing now, where the exposure is real and nobody has been told it is theirs. Most of the mistakes and the failures and the public embarrassments between here and there will happen for that one reason.

Which points at a better question, and it's one you can ask this week rather than in three years.

If that role existed in your organisation tomorrow, whose name would be on it?

Ask it and watch what happens. Most people go quiet. Then they say a name. Then they say, "but I don't think they know."

### What this actually means for you on Monday

You aren't going to read any of that. Nor should you.

Here is what transfers, in three lines.

**The law is not converging and waiting for clarity is not a strategy.** There is no moment coming where this settles and somebody circulates the summary. The organisations that do well will be the ones that built a defensible way of working before anybody made them.

**Contract is the main tool, precisely because the substantive law is unsettled.** When nobody is certain who is liable, the allocation that matters is the one you wrote down with your supplier. Warranties, indemnities, what happens on a breach, who carries what. That isn't legal exotica. That is a procurement conversation, and procurement conversations are had by ordinary managers every week.

**Adopting a recognised framework is not box-ticking. In at least one jurisdiction it is literally a defence.** There is an American state statute under which adopting a specific national risk-management framework functions as an **affirmative defence**. The framework is voluntary. Using it's a legal shield. Almost nobody in a non-technical role knows that sentence exists, and it is the single most useful thing you can say in a governance meeting.

### The eleven questions

None of these is a legal question. Every one can be asked by somebody with no legal training, and asking three of them in the right meeting will change how you're seen for a year.

**1. Are we the builder or the user of this, and do we know which obligations follow from that?** The answer is almost always *user*, and almost nobody has checked what that means. It is the question the rest depend on.

**2. Does anything we use touch a decision about a person?** Hiring, promotion, pay, credit, insurance, housing, education, access to a service, or who gets contacted and who doesn't. That is the boundary that turns ordinary software into the regulated category, in every regime, everywhere. If the answer is yes anywhere in your business, everything else on this list becomes urgent.

**3. Whose law reaches us?** Not where your office is. Where your **customers**, your **staff** and your **data** are. A company in one country with users in another is subject to the second one's rules, and the map of your obligations looks nothing like the map of your buildings.

**4. Can we show a human was involved, and that they could actually have changed the outcome?** Human oversight is not a person on an org chart. If the human sees the output after it has gone, or has no realistic ability to override it, the oversight is decorative and will be read as decorative.

**5. Do we tell people?** When someone is dealing with an AI rather than a person, when content was machine-generated, when a decision about them was substantially automated. Transparency duties are the most common feature across every regime I've looked at, and the cheapest to comply with, and the most commonly missed.

**6. Have we adopted a recognised framework, and can we evidence it?** Not *do we have a policy.* Can we show the assessments, the decisions and the dates. A framework you adopted and never evidenced is worth nothing when it matters, which is the exact moment it was for.

**7. What do our contracts say about who carries this?** With every vendor whose system touches a decision. If nobody can answer, the answer is you.

**8. If we had to produce the record, could we?** Which system, which version, which question, which output, who reviewed it, when. Not because a regulator will definitely ask. Because you can't defend a decision you can't reconstruct, and reconstruction is not something you can retrofit after the letter arrives.

**9. Who signed this off, and do they know they did?** In most organisations, a tool arrived, somebody sensible started using it, it spread, and no one ever made a decision. There is a name at the top of an accountability chain and quite often that person has never been told.

**10. What is our plan for the day this changes?** Not if. The volume of legislation above means something relevant to you will change inside twelve months. Is there a person whose job includes noticing, or is your plan to read about it in the press?

**11. If this went wrong publicly tomorrow, what would the first line of the news story be?**

That last one is the one to ask out loud.

It is not a legal question at all and it does more work than the other ten combined, because everyone in the room can answer it instantly and nobody can pretend they can't. *Company uses AI to reject applicants and can't explain how.* *Firm's chatbot gives wrong advice for six months.* *Staff pasted client records into a public tool.*

If the sentence comes easily, you've found the thing to fix. And you found it in a meeting, using no expertise, at a cost of about four seconds.

### What question two turns up when you actually ask it

I have given you eleven questions and no picture of what happens when one gets asked, so here is question two doing its work. *Does anything we use touch a decision about a person?*

My prediction is that the first answer you get will be no. It will come back quickly and confidently, from somebody senior enough that everybody else in the room relaxes — no AI in HR, no algorithmic hiring, nothing anywhere that decides anything about anybody.

Then somebody near the end of the table says: what about the sift?

The company had been getting more applications than it could read. So a manager · not IT, not HR, a manager with a problem and a deadline · had started pasting batches of CVs into a general-purpose tool with an instruction that amounted to *which five of these best match this job description.* She then read those five properly. She was doing it in good faith, she was doing it on top of her actual job, and she had told her own director, who thought it sounded sensible and efficient, because it is.

Now walk it through the list.

**Question two.** It touches a decision about a person. Hiring is named explicitly in every regime anybody has written.

**Question one.** They are the deployer, not the builder. The obligations that follow are theirs, and the tool's terms of service protect the tool's maker.

**Question four.** Can they show a human was involved and could have changed the outcome? A human read the five. Nobody read the other sixty, and the five were chosen by something nobody in the building could explain. The oversight is real and it sits entirely on the wrong side of the filter.

**Question eight.** Could they reconstruct it? Which version answered, what the instruction said that week, which sixty were rejected and why. No. It happened in a chat window on one person's account.

**Question nine.** Who signed this off? Nobody. A manager solved a problem she had been left with, and every person above her who heard about it thought it sounded efficient.

Five of the eleven, and not one of them needed a lawyer to ask. The whole thing surfaced in under four minutes, in a meeting that had opened with a confident no.

**And now the part I care about more than the finding.** Nobody in that story did anything wrong in the ordinary sense. I do not think there is a villain in it, and I have never found one in any version of it I have come across. There is a person who was handed more applications than a person can read, was handed nothing else, and reached for the tool her whole industry reached for that year.

That is what this exposure looks like from the inside, and it is why I keep saying the question has to be asked out loud by somebody. It will never announce itself. **It looks like somebody coping.**

### Why this makes you look wise

Be direct about the mechanics, because that's what this book is for.

Senior people are genuinely uneasy about this and almost nobody brings it to them in a form they can act on. What reaches them is either a vendor selling reassurance or a technical briefing about how the model works, and neither answers the question they are actually holding, which is *what happens to us if this goes wrong.*

Walk in with question two and question eleven. *Does anything we use touch a decision about a person? And if it went wrong publicly tomorrow, what is the first line of the story?*

You have claimed no expertise. You do not need any. You have asked the two questions the room has been circling for a year, and you'll be in the next conversation about it, which is a room you were not previously in.

There is a version of this that goes wrong, and I want to name it so you avoid it. Do not walk in as the person who has found a problem. Walk in as the person who has found the **question**, and let them own the answer. The first is a threat. The second is help.

### One thing to do this week

Take question two. *Does anything we use touch a decision about a person?*

Ask it about your own team first, quietly, before you ask it in a meeting. Hiring, or who gets contacted, or who gets flagged, or how work is allocated. You may find nothing. You may find something that has been running for eight months.

Then, whichever it's, you know something about your own organisation that almost nobody else in it knows.

Which is a useful position to be in, and a completely useless one, until somebody other than you can see it.


## 16 · Make It Visible, Then Give It Away

Correct work that nobody can act on isn't half a win. It is indistinguishable from not having done it.

Watch somebody the moment they understand something.

They pause. Their breathing changes. Their head tilts slightly, as though the information arrived from an angle they were not facing. And then some version of the word *wow* comes out of them, usually quietly, often to nobody.

I have watched that happen a lot now, and I can tell you exactly what it isn't. It isn't the moment they received good information. They received that a few seconds earlier and it did nothing.

It is the moment they **could see it.**

That gap, between correct and visible, is where most careers quietly stall. You did the work. You checked it, you scored it, you used two systems and read the disagreement. Then you put it in nine paragraphs of prose and sent it at 6pm, and the person who had to act on it skimmed the first two and made the decision they were already going to make.

Which is the sentence at the top of this chapter, arriving on your own work.

### The first job is to be understood, not to be right

Your instinct, once you get good at the checking, is to add more. More caveats, more context, more of the reasoning that made you confident; you are proud of the reasoning. It cost you something.

The person receiving it does not want your reasoning. They want to run their own decision on top of your work, and every extra paragraph is a tax on their ability to do that.

So the last pass on anything isn't another check; it is: **show me this differently.**

Make it a table. Make it a chart. Make it five bullets and a recommendation. Explain it for somebody with ninety seconds who has to choose.

And do that with the same tool you did everything else with, because it's genuinely good at this and it takes one sentence. *Give me this as a table with the decision in the left column.* *Turn this into something a person could read in sixty seconds.* *What is the one chart that would make this obvious?*

You will find the answer was often a table all along, and you had written it as prose because prose is what you were taught to produce.

### The same finding, twice

Here is what that one sentence does. The same work reaching the same conclusion, written by the same person, ninety seconds apart.

**What you wrote:**

> Following our review of the three shortlisted suppliers, we have considered pricing, implementation timelines and support arrangements. Supplier A offers the lowest headline cost but their implementation window of sixteen weeks would take us past the Q3 deadline, and their support is business-hours only. Supplier B is approximately 12 per cent more expensive but can implement within nine weeks and offers extended support hours, although their contract terms include an annual uplift clause which is currently uncapped. Supplier C sits between the two on cost and timeline but has the strongest support offering, and is the only one of the three able to provide a named account manager. On balance we would suggest that Supplier C represents the most appropriate option, subject to further clarification of their pricing structure.
>
> *(129 words. Everything in it is true.)*

**What you send:**

> **Recommendation: Supplier C** · the only one that hits Q3 with support we can hold somebody to.
>
> | Verdict | | Cost | Live by | The one thing to watch |
> |---|---|---|---|---|
> | **Take it** | **C** | +6% | Week 10 | Pricing structure unclear. One question, answer by Friday |
> | Rule out | A | Lowest | **Week 16 · misses Q3** | The timeline kills it. Nothing else about A matters |
> | Hold as second | B | +12% | Week 9 | **Uncapped annual uplift** |
>
> *All three costs and A's and B's dates are taken from the quotes. C's week 10 is their own estimate and I have not verified it · if it slips a fortnight we are where A already is.*
>
> **If you want A anyway:** the Q3 date moves. That is the trade.

Read the two again and notice what actually changed, because it was not the writing.

**The prose hides the decision inside a paragraph of considerations.** The table puts the thing that kills supplier A in bold, in the column where it belongs, where you cannot read past it. *On balance we would suggest* became **Recommendation: Supplier C**, and the uncapped uplift stops being a subordinate clause buried thirty words into a sentence about something else — which is where the second most dangerous fact on the page had been sitting.

**And the last line is the one that gets you invited back.** *If you want A anyway, the Q3 date moves.* You have pre-answered the pushback and handed the decision back to the person whose decision it is, without pretending you do not have a view. Nothing in the prose version does that, and it is not because the writer did not know it. It is because prose has nowhere to put it.

Identical facts, and about ninety seconds of extra work.

### People do not receive the same way

There is a thing every manager knows in principle and forgets in practice. **People take information in differently.**

Some will listen to a five-minute summary before a meeting and arrive fully briefed. Some want a page they can read and mark up. Some won't understand anything until they've sat with the thing themselves and turned it over.

You aren't going to know which one you've got, and asking is awkward.

So the version you produce should survive all three. A short spoken summary, a page that stands alone, and the underlying detail available for the person who wants to dig. That sounds like three times the work. It is one extra instruction, because the material already exists and you're only asking for it in another shape.

The cost is about ninety seconds. The difference in whether it lands is total.

### Show where every number came from

Now the part that builds something you cannot buy, which is a reputation for being reliable.

On client work, every figure we publish carries a colour. **Green** means verified against the primary source. **Amber** means desk research, plausible, unverified. **Grey** means open, we don't know.

Three colours, next to the numbers, on the face of the document.

One of the tabs on a recent piece of work is **entirely amber**, and it says so at the top.

Read that again, because it is the whole idea. A page of work that announces, before you read a word of it, that none of it has been verified.

Every instinct says that's a bad look. It is the opposite. The client now knows exactly which parts they can take to their board and which parts need another week, and they learned that in one second instead of finding out in the meeting.

And here is the thing that surprised me. The amber page made the green pages **more** believable, not less. If you're willing to mark your own work amber, your green means something. A document where everything is presented with equal confidence tells the reader nothing about any of it, and a careful reader knows that, so they discount all of it evenly.

You don't need colours. You need the habit. *This figure is from the filing. This one is an estimate. This one I couldn't confirm.* Say it in the document, not in the meeting when somebody asks.

That sentence is what your senior people have been waiting years for somebody to say.

### Check it renders before you admire it

One practical thing that will otherwise undo all of the above.

Charts sent as inline data, embedded directly in the message, are the technically elegant way to do it. They also fail to render in some of the most widely used mail clients, which means a proportion of your readers open a considered visual and see a blank space or a broken icon, and draw the obvious conclusion about the person who sent it.

Attach them as ordinary image files instead. Duller, and it works everywhere.

The general rule is worth more than the specific fix. **A visual that does not display is worse than no visual**, because it's not neutral. It is evidence of carelessness, arriving in place of your argument.

So look at it. Send it to yourself. Open it on a phone. That is thirty seconds against the alternative, which is a senior person's first impression of your work being a grey box with a question mark in it.

### Then give it away

Here is the part people resist, and it's the highest-return thing in this chapter.

When you work out how to do something well, **give away the method.**

Not the output. The method. The exact instruction you used, the check that caught the error, the way you got it into a table that made the decision obvious. Send it to the person who has the same problem. Put it where your team can find it. Show somebody in ten minutes what took you two weeks.

The instinct against this is real and I understand it. If the method is what makes you valuable, giving it away spends the thing that makes you valuable.

That instinct is wrong, and it's wrong for a specific reason.

**The method is not the scarce thing.** It is a paragraph. Anyone can copy it and most people won't use it, because the hard part was never the instruction, it was the judgement about when to apply it and the discipline to act on a bad score.

What you actually get for giving it away is this. You become the person who made three other people better at their job, and that's a fundamentally different thing to be than the person who produced good work. One of them is an employee with strong output. The other is somebody whose departure would cost the organisation something it cannot easily replace.

There is also a quieter effect. Teaching it forces you to understand it. The first time you explain your checking process to somebody who asks why, you'll find the two steps you do out of habit and cannot justify. Keep one. Drop the other.

### What giving it away actually looks like

People hear *give away the method* and picture running a session, so let me be exact about the size of this, because the version that works is much smaller than the version people imagine and never happens.

It is one message, to one person, about one thing they already have a problem with.

> *You mentioned the board pack takes you two days. This is the instruction I use on mine, paste it under whatever you've drafted. The bit that does the work is the last line · without it you get a lecture instead of a question.*
>
> *Score this out of 100 across five dimensions: hook, audience, proof, emotion, call to action. 95 and above is exceptional and rare, most first drafts land between 76 and 88, score accordingly. For each one give the score, one sentence on why, and one fix. Then rank the fixes by how many points each would gain. For anything under 82, give me one question that would fix it, maximum twenty words, no preamble.*

That is it. That is the whole act of generosity, and it takes about ninety seconds to send.

Now watch the three things that come back, because none of them is the one people expect.

**The first is that they use it and it does not work the way it did for you**, and they tell you so. They ran it on a technical document and the *emotion* dimension came back as noise, because nobody has a feeling about a network diagram. That is not them failing. **That is your method being tested against a case you did not have**, and the correct response is to change the dimensions for that kind of work, which you now know to do and would never have found alone.

**The second is that they add something.** They started putting the threshold at the top of the instruction rather than in their head, because they kept forgetting it. That is better than what you were doing. Take it.

**And the third is the one that changes your position.** Somebody they work with asks where they got it. Your method is now travelling without you in the room, attached to your name, doing work in a part of the building you have never been in. You did not do anything to make that happen and you could not have made it happen deliberately.

**The version that fails**, and I want to name it so you avoid it, is the one where you explain the whole system. You send four paragraphs of context about why checking matters, the person reads none of it, and nothing changes. Give one technique, on their own real task, with the actual words to type. Not a description of the words. The words.

Somebody who can copy and paste something in ninety seconds will try it. Somebody who has to understand your philosophy first will mean to try it, and will not.

### The trap in all of this

The common failure here is not producing too little. It is producing a great deal and never asking whether any of it worked.

Charts, one-pagers, colour-coded tables, provenance markings · a year of them, and no record anywhere of whether a single one made a team decide faster, or better, or at all.

**Volume is easy to count and impact is not**, so people count the one that is easy and call it evidence. It is not evidence. It is activity.

The whole correction is one habit: when something you made visibly changed a decision, write down what it was. Three lines a month. In a year you will have the thing almost nobody has, which is a record of your own work landing.

### What to do with the next thing you finish

Three passes, and the whole thing takes under five minutes.

**Show it differently.** *Give me this as a table, with the decision in the left column.* Read both versions and keep the one that a busy person could act on.

**Mark your confidence.** Go through the numbers and label them. Verified, estimated, unknown. Put the labels in the document. If a whole section is unverified, say so at the top of it.

**Give the method to one person.** Not the output. The instruction that produced it, and the check that caught something.

Do those three on your next piece of work and something will happen that has probably not happened to you before. Somebody will act on your work in the meeting rather than taking it away to think about.

That is the difference. Not being right more often. Being **usable**.

And once that starts happening regularly, once you're the person whose work gets acted on and whose method other people are using, you'll have a problem you've never had before.

---

> ### ▪ DO THIS
>
> **Four passes on the next consequential thing you finish.**
>
> **1. Show it differently.** *Give me this as a table, with the decision in the left column.* Read both versions. Keep the one a busy person could act on.
>
> **2. Mark every number.** Verified, estimated, unknown. Put the labels in the document, not in your head.
>
> **3. If a whole section is unverified, say so at the top of it.** This is the step that feels wrong and is worth the most.
>
> **4. Give the method to one person.** Not the output. The instruction you typed, and the check that caught something.
>
> **Under five minutes if you noted your sources as you went. If you didn't, pass 2 is the work. Budget an hour the first time.**
>
> **When marking your own work unverified feels like weakness:** it does the opposite. A document that admits which parts are soft makes its firm parts believable. One that presents everything with equal confidence tells the reader nothing about any of it, and careful readers discount all of it evenly.
>
> **The signal to watch for:** people acting on your work in the meeting rather than taking it away. Treat that as the sign it landed, and start keeping a record of when it happens.

---

You will have evidence. And you'll have to decide what to do with it, because nobody is going to walk up and offer you the conversation.


## 17 · Ten Days

Ten days. One job a day. Every one is built to fit inside thirty minutes and every one is done on work you already have to do. Days eight and nine depend on somebody else's diary, so those two can run longer, and that's the only reason any of them will.

Do not read them all now. Read day one, do day one, come back tomorrow.

I spent thousands of hours getting to what is in these ten days. Fourteen and sixteen hour days, seven days a week, for months, because nobody could tell me the order to learn it in and I had no way to find out except by doing all of it badly first. That isn't a boast and you should not copy it. It is the reason this chapter is thirty minutes.

Everything below is what survived. The order isn't arbitrary either; it is the order in which each thing stops being hard, which is not the order I learnt them in and is the single most useful thing I can hand you.

### You finish holding something

This is the part that matters, and it's why the days are in this sequence rather than any other.

Each day produces one artefact you keep. By day ten you're holding all of them, and together they aren't ten exercises. They are the evidence.

| Day | What you keep |
|---|---|
| 1 | A written brief you will reuse for a year |
| 2 | Proof the first answer was not the best one available |
| 3 | One instance of the feeling, in your own handwriting |
| 4 | A claim that survived an attack, or did not |
| 5 | A number, on work you had already called finished |
| 6 | One disagreement you would never have found alone |
| 7 | One honest answer about your own exposure |
| 8 | Work that got acted on in the meeting instead of taken away |
| 9 | One other person doing something they could not do last week |
| 10 | Three sentences you could say out loud in a review |

Look at that list again as a single object. A reusable method, a before and after, a piece of your own work scored honestly, a risk you named before anybody asked, and someone else who is better because of you.

That isn't a training log. **That is the case, and the next chapter is the conversation where you put it on the table.** You cannot assemble it the night before. It takes ten days, which is why they're here and not in an appendix.

If you read the last nine chapters passively, nodding, without ever opening the tool and trying it, this chapter is where that catches up with you. That isn't a scolding. It is just how it works. Nobody has ever got better at anything by agreeing with it.

### Before day one · somewhere to be bad at this

Ten days of this on live work, with your name on every output, is a very good way to make yourself careful. Careful is the wrong setting for learning. So don't start there.

The failure I expect for most people reading this is being told to press one button, at one time, in the approved way. Do that for a year and you have a dependency instead of a skill, and nothing anybody would pay extra for. You are operating a lever.

What you want is a sandbox. In a technical team that means somewhere you can press everything and break nothing. You probably can't requisition one of those, and here is the good news: you don't have to ask anybody, because you already own the ingredients.

Pick work that is genuinely yours and genuinely finished. Last quarter's report, filed. A decision already made and gone. An email you already sent. Nothing live, nothing confidential, nothing anybody is waiting on. That is your sandbox and you can assemble it this afternoon without a single conversation.

Then use it the way pilots use a simulator. They do not practise on passengers. They fly the failure, over and over, in a room where the failure costs nothing, which is exactly why they are calm on the day it stops being a simulation.

Go back to it every few months. These systems change underneath you, the model behind the tool you use will be replaced while you aren't looking, and the thing that didn't work in March will work in June. If the only place you ever meet the new version is live work, you will meet it badly.

---

### Day 1 · Write one brief

**Your outcome: one instruction that a stranger could follow without asking you a question.**

1. Pick a task you already hate. Something you do weekly and resent.

2. Write the six lines from chapter four before you write the request. Identity, who it should act as. Goal, what outcome for whom by when. Good, what a finished one looks like. Bad, what it must not do. Challenge, what it should argue against itself. Proof, what counts as evidence.

3. Now write the request underneath those six lines and send the whole thing.

4. Compare what comes back with what you usually get.

Most people have never written the six lines. They have written the request a hundred times and wondered why the answer keeps arriving generic.

**The part that matters most sits inside line two: who the outcome is for.** Not the topic, the person. Their role, what they already know, what would waste their time. It changes the vocabulary, the length, the sophistication and what gets left out, and it costs you nine words.

Keep the six lines. You will reuse them tomorrow.

---

### Day 2 · Ask a different question

**Your outcome: proof that the first answer was not the best available one.**

1. Take yesterday's output. Do not ask whether it's correct.

2. Ask instead: *what is this missing that the reader would notice?*

3. Then: *show me the same thing organised completely differently.*

4. Then: *what would make this useful rather than just accurate?*

5. Put the first answer and the fourth side by side.

Three asks. Under four minutes. Notice that none of them was *are you sure* and none of them was about accuracy.

The gap between those two versions is what you have been leaving on the table every time you accepted the first thing that appeared. Look at it properly before you move on, because tomorrow depends on you believing it.

---

### Day 3 · Catch yourself

**Your outcome: one instance of the feeling, written down.**

Today you don't produce anything. You watch.

1. Work normally. Use the tools as you always do.

2. Every time you notice the sentence in your head. *I hope this is right*, *I didn't know that*, *that's not what I expected*. Write down what you were looking at. One line. A note on your phone is fine.

3. At the end of the day, count them.

Most people get between two and six. Some get none on day one and four on day two, once they know what they're listening for.

You have had that signal your whole career and treated it as noise. Today it becomes data. Tomorrow you act on it.

---

### Day 4 · Make it prove one thing

**Your outcome: a claim that survived an attack, or did not.**

1. Take one thing from yesterday's list. Just one, the one that nagged most.

2. Type: *use an adversarial agent. Give me the sources as clickable links, in a grid, one row per claim, with a certainty score from 0 to 100 on each row.*

3. Look only at the rows scoring under 60.

4. Chase one of them to its actual source.

Time it. It will be about six minutes.

You will find one of two things. Either something was wrong, and you've just saved yourself. Or nothing was wrong, and you now have the right to put your name on it, which you didn't have yesterday.

Both outcomes are wins. Only one of them feels like one, and that's worth knowing in advance.

---

### Day 5 · Score something you already liked

**Your outcome: a number under your own threshold, on work you thought was finished.**

1. Before you do anything else, write down your bar. *This doesn't go out below 92.* Write it where you can see it. Chapter eight explains why the bar goes above the normal range rather than inside it.

2. Take a piece of work you have already decided is fine.

3. Ask: *score this out of 100 across five named dimensions. For each, give the score, one sentence on why, and the single change that would raise it most. 95 and above is exceptional and rare. Most first drafts land between 76 and 88. Score accordingly.*

4. Read the lowest-scoring dimension's reason.

5. Make that one change.

**Those numbers are mine, not a standard.** Seventy-six to eighty-eight is where my first drafts land and ninety-two is where I set my bar. If your work has a different normal, use yours. What matters is that a range is stated at all, because without one it hands you ninety-one every time.

The threshold goes first for a reason. If you look at the score before you set the bar, you'll set the bar just underneath the score, and you'll do that every time without noticing.

The uncomfortable part is step five, on a piece of work you had already finished. Do it anyway. That is the whole day.

---

### Day 6 · Use a second one

**Your outcome: one disagreement you would never have found.**

1. Take a decision that actually matters this week. Not a task, a decision.

2. Put the same question into two different tools. Different companies, not the same one twice.

3. Ignore both answers for a moment.

4. Find where they part company. Write down that one point.

5. Spend your attention only there.

Four minutes. The gap is the map, and it points at the single paragraph where the difficulty actually lives.

While you're there, sort yesterday's tasks into judgement and typing. Put the typing on the cheaper, faster option tomorrow and see whether anything gets worse. Usually nothing does, and that's the day's second lesson arriving free.

---

### Day 7 · Ask what could go wrong

**Your outcome: one honest answer about your own exposure.**

1. Ask yourself, before you ask anybody else: **what is the most sensitive thing I have personally typed into one of these tools?**

You will remember something. Everybody does. Sit with it for a second rather than moving on.

2. Now go and look at two settings on the tool you use most. Whether your inputs are used for training. How long they're kept.

3. Take one thing you were about to type today and rewrite it so the identifying part isn't there. The name becomes *a long-standing client*. The number becomes *a six-figure invoice*.

4. Send both versions. Compare the outputs.

They will be the same. That is the point and you have to see it yourself to believe it, because the identifying details feel essential right up until you remove them and nothing happens.

---

### Day 8 · Make one thing visible

**Your outcome: work that gets acted on in the meeting rather than taken away.**

1. Take something you've finished and would normally send as prose.

2. Ask: *give me this as a table with the decision in the left column.* Read both. Keep the better one.

3. Go through every number in it and mark where it came from. Verified. Estimated. Unknown.

4. If a whole section is unverified, say so at the top of that section.

5. Send it.

Step four is the one that feels wrong and is worth the most. A document that admits which parts are soft makes its firm parts believable. A document that presents everything with equal confidence tells the reader nothing about any of it.

Watch what happens in the meeting. Somebody will act on it instead of asking for more time.

---

### Day 9 · Give the method away

**Your outcome: one other person doing something they could not do last week.**

1. Pick the single most useful thing you've learned in the last eight days. Probably day 4 or day 5.

2. Find one person with the same problem.

3. Show them in ten minutes. Give them the actual instruction you typed, not a description of it.

4. Do not explain the whole system. One technique, working, on their own real task.

The instinct against this is that the method is what makes you valuable. It is not. The method is a paragraph. The judgement about when to use it's the valuable part, and that doesn't transfer in ten minutes.

What you get instead is a change in what you're. Not an employee with strong output. Somebody who made three other people better, which is a different category and is priced differently.

---

### Day 10 · Write down what changed

**Your outcome: three sentences you could say out loud in a review.**

Today you produce the only artefact that matters for what comes next.

1. Look back over nine days and find **one thing you caught** that would otherwise have gone out. Write one sentence. What it was, what it would have cost.

2. Find **one thing you produced faster or better** than you could have two weeks ago. Write one sentence. Be specific about the before and the after.

3. Find **one person you made more capable.** Write one sentence.

4. Read the three sentences out loud.

That is it. That is the whole ten days, and it fits on an index card.

Those three sentences aren't a diary entry. They are the evidence, in the form the only conversation that matters requires, and almost nobody walks into that conversation holding anything at all. They walk in with a feeling that they've been working hard, and a feeling isn't something the person opposite can act on.

You now have something they can act on.

---

### If you did not do it

Be honest with yourself, because nobody else is going to check.

If you read these ten days without doing them, you've read a description of a method rather than acquiring one, and the difference will show up in about six months when somebody who did the work is doing things you can't.

Go back to day one. It is thirty minutes and it's on a task you have to do anyway.

The only thing standing between you and the last chapter of this book is ten short days of doing rather than agreeing, and the person who decides what you're paid doesn't know or care which one of those you chose; they will only see the output.

Which brings us to them, and to the conversation you're now, for the first time, actually equipped to have.


## 18 · The Conversation That Changes Your Income

The best pay conversations I've ever had lasted about four minutes.

Not because I'm generous. Because by the time the person sat down, there was nothing left for either of us to decide. They already knew, I already knew, and the meeting was where we agreed on a number and got back to work.

That is the good news in this chapter, and I want you to have it before anything else.

The conversation is the easy part. What decides your income happens months earlier, in ordinary weeks, in small moves nobody schedules. Every one of those moves is entirely within your control, which is more than can be said for almost anything else about your career.

I have run businesses and advised many more, which means I've spent years in the chair where somebody asks and I decide. Almost every book that tells you how to have this conversation is written by somebody who has only ever asked.

So here is what it looks like from the other side, and how to make it a four-minute meeting.

### What I am actually buying

You think you're asking me for money. What I'm buying is certainty.

That word does more work than anything else in this chapter. A business runs on predictability.

Somebody who takes doubt out of an area of it is worth real money. When I'm confident a thing will simply operate well because they're in it, that confidence has a price and I will pay it happily.

Notice what that means for you. You aren't competing on being clever, or liked, or hard-working. You are competing on how much doubt you remove.

That is a much easier game than the one most people think they are playing, and it's winnable inside six months.

### The five things that get a yes

When I have said yes, and I've said it many times, it has been for some combination of these. Read them as a list of things to build, because that's what they are.

They produce results I couldn't easily replace. Not effort and not hours. Results of a kind I would struggle to buy elsewhere at the same price.

Something would visibly get worse without them. I can name the thing. So can they. That is the difference between being valued and being valuable.

They give me certainty, as above, and it's the most underrated item on this list.

Somebody else wants them. When another organisation is trying to attract somebody, my decision gets very easy. On one side is more risk, a great deal of challenge, and possibly failure. On the other is certainty. Businesses will pay for certainty, and that's most of what they are buying.

They make the people around them better. Somebody who works well with others and helps them get more productive is worth more than their own output, and every manager alive knows it.

Look at what is not on that list. Loyalty. Time served. Working late. Being good company.

I value all four of those, and not one of them has ever moved a number for me when it came to deciding what to pay somebody. They moved a different decision entirely, which was whether I want to keep them.

Separate those two in your head, because almost everybody confuses them and spends years being excellent at the wrong one.

Loyalty keeps you. It won't get you the rise.

### The arithmetic that makes it easy

Here is the sum from my side of the desk, and it's simpler than most people asking assume.

Giving somebody twenty per cent more is straightforward when their output has gone up by a hundred. That isn't generosity. It is obvious arithmetic, and any manager who can see the output will run it in their head in about four seconds.

The mistake almost everybody makes is asking me to accept the twenty without ever showing me the hundred.

There is a second thing happening at the same time, and it works in your favour without your doing anything about it.

Businesses actively try to de-risk their unique people. When somebody holds skills and knowledge we can't easily replace, keeping them stops being a cost question and becomes an insurance question, and those are decided in a different part of the brain and usually with a different budget.

Think about why nobody haggles with the one specialist who can actually do the thing. It isn't that they work harder than everyone else. It's that the alternative is a search, a gap, a risk, and a stretch where nobody knows how it will go.

That is the position you're building towards. Not indispensable through hoarding, which fails, but hard to replace because of what you can actually do and how few people around you can do it.

### The constraint I am under, and how you help me with it

Here is something almost nobody knows, and knowing it will change how you ask.

If I increase one person, others will ask. They find out, or they sense it, and within a quarter I'm having four versions of this conversation instead of one.

So the question in my head is never quite "does this person deserve more." It is "can I explain this one to everybody else."

That sounds like an obstacle. It is actually the most useful thing in this chapter, because it tells you exactly what to give me.

Give me the sentence I can repeat when you aren't in the room. Something short, factual and specific about what you now do that nobody else does. If you hand me that sentence, you haven't asked me for a favour; you have handed me the argument, and I will use it.

Most people spend the meeting explaining how hard they work. Almost nobody hands over the one line that makes the decision defensible.

### What I watch others in my chair get wrong

I'll tell you what is happening in rooms like mine, because it sets what you are worth.

Tens of thousands of people are being laid off, and small companies are doing it too, because they have put AI into something and the need has gone. That is real and I am not going to soften it for you.

What I think about most of those decisions is that they were made far too early.

A company that cuts instead of retraining is usually doing one of three things. Reaching for the bottom line because a shareholder is watching. Taking the profit. Or quietly solving a different problem, which is that it has people who will not change and it has not wanted the conversation.

Not one of those is a judgement about whether the work still needs a person.

And the sequence that follows would be funny if you weren't standing in it. They fire. Then they discover they have to rehire. Then they have to retrain whoever they hired. A cycle, paid for twice, arriving back where it started with less knowledge in the building than it had at the beginning.

Here is what I would rather do, and what I think most people in my chair would say privately. I have somebody who has given me years of their life. Spending some of my money to take them somewhere new is a better trade than replacing them, and on the occasions it works it is the best thing that happens all year. Somebody who takes to this properly does not simply do their old job faster. They take the business somewhere it was not going.

Which is the whole reason this conversation is worth having. You are not asking me to be generous. You are showing me you are the one worth spending it on, and that I do not have to run that cycle.

### The six months that actually decide it

This is the part that turns a difficult conversation into a four-minute one.

Tell the business what you're working on, as you go. Not as a boast. As a running commentary, in ordinary conversations, that you are building AI strategies, learning the techniques, and applying them here.

They need to trust that this is something you genuinely care about, know about, and have been doing for a while. That trust takes months, and it can't be manufactured in a meeting.

What you're really doing is bringing people on the journey with you. When they've watched you learn something over half a year, three things happen.

They have confidence you'll keep going, because they have seen you keep going. They connect the results to the effort, because they saw both. And when you finally ask, nothing is a surprise, so nobody has to make a decision under pressure.

Do it in small moments. Mention what you tried and what didn't work. Send the thing you built to one person who will find it useful. Tell your manager once, plainly, that you're working on this and why, and then let them watch.

Six months of that and you aren't asking. You are confirming.

### Create the moment

The other thing almost nobody does: you have to make the moment happen.

Businesses don't naturally produce decision points. They produce financial years, appraisal cycles and reorganisations, none of which are designed to notice you.

Waiting to be noticed is the most common career mistake I see, and it happens most often to the people doing the best work.

So you create it, and you lay it out logically.

By then I'll already know you have the skills. What I'll not have done is assemble the pieces into a single case with a number attached, because that was never my job; it is yours, and it's the last mile of everything else in this book.

Bring the before and the after. Bring what it cost and what it produced. Bring it in my units, which are money, time, risk and headcount.

### What that actually looks like on my desk

I have spent this chapter telling you to bring me something I can act on, and I have not once shown you what it looks like, which is exactly the failure this book keeps catching in everybody else's work. So here is the whole thing.

It is one page. Every line on it comes out of the ten days in the last chapter, and the person who wrote it spent about twenty minutes assembling it from notes they already had.

> **What changed, and what it is worth**
> Prepared for the March review
>
> **What I took on.** Nobody asked me to do either of these. I rebuilt how we produce the monthly supplier pack, and I now run the exception review that used to come to you.
>
> | | Before | Now |
> |---|---|---|
> | Monthly supplier pack | two days | three hours |
> | Exception review | came to you | I do it · you see a summary |
> | Errors caught before send | not counted | four since January |
>
> **What it cost.** About twenty minutes a day for ten days, on work I was already doing.
>
> **The one that mattered.** In February the pack carried a supplier rate that had rolled forward from a contract which expired in November. It was going to the board. I found it because I now check every figure against its own source rather than against last month's pack, and last month's pack was where the error had been living since December.
>
> **Who else can do this now.** Two people on my team run the same check on their own packs. Neither of them asked me · they watched it work and wanted it.
>
> **What I want.** To own the supplier pack end to end, including the exception decision. A title that says so. And a number that reflects both.

Read that again and notice what is not in it. No adjectives. Nothing about how hard the ten days were, nothing about being committed or proactive or passionate, and not one sentence asking me to be pleased.

It is four facts and a request, and I can carry every one of them into the room where I have to justify it, which is the room you are never in and the only one that decides anything.

The February line is the one that does the most work, and it is worth understanding why, because most people would have left it out for being too small. It is not the size of the error. It is that the sentence contains a thing that would have happened and did not, and I can picture it. Two days becoming three hours is a number I have to take on trust. A rate that expired in November arriving in front of the board in February is a morning I can see myself having.

Give me one of those and the rest of the page stops being a claim about you and becomes a description of something that already happened.

### Ask for the right thing

Most people ask for a salary number and stop there, which leaves the best part on the table.

Ask for more to control. An area to own. More people, or more machines, and increasingly the second is the more interesting request. Somebody who says I want to own how we run this, and I'll be accountable for the outcome is asking for something a business finds far easier to give than money.

It is also worth considerably more to them within two years.

Ask about the title. This is the piece of inside information I would most want you to have.

Titles set your wage category, and businesses are generally flexible about a name change. A new title costs the company almost nothing today, and it moves the band you sit in for the rest of your time there and for the job after this one.

People spend the whole conversation fighting over a percentage and never once mention the word on the door.

On the number, be specific. Know that specific work leads to a specific number. If other companies are asking you to do that role, that market rate flows naturally into what you can be paid here.

Which brings us to the part that needs care.

### Leverage, handled well

Being wanted elsewhere is real leverage and it's the fastest route to a yes.

It is also the one thing that can cost you everything if you handle it clumsily. If it looks like you're trying to leave, the company stops seeing you as loyal, and once that shifts you rarely get it back; you can win the money and lose the future.

So keep the balance. Build the skills. Have the conversations. Take on more. Then, at the point where you're demonstrably doing more, talk about what that "more" should mean for you.

Do not say the words "I have another offer" unless you're genuinely prepared to accept it. Your leverage is that I can see what you are worth, not that you've told me somebody else can.

One thing to avoid entirely, and I mention it only because it's the easiest way to turn a yes into a no.

Never build your case on what other people earn. It is the one argument that has nothing to do with your own value, and it leaves me rewarding something I can't repeat.

I know what that costs, because I watched it cost somebody.

A technical manager in a health business asked me to move them from $100,000 to $120,000. They were worth it. I want to be clear about that before anything else, because it is the part that matters.

They did not get it.

They had found a way to see what other people were paid, and they came in holding it. Not as evidence. As pressure. What I heard underneath was *give me this, or else*, and once I had heard it I couldn't unhear it.

Because I was not deciding about $20,000 any more. I was deciding whether somebody who had just shown me they would use what they knew against the company should keep running the systems I couldn't check myself. That is a different question and it only has one answer.

Knowing something does not make it yours to use. That is true of a salary spreadsheet, and it's about to be true of a great deal more, because you are all about to be able to find out things you couldn't find out before.

**And now the part that was mine.** They were worth $120,000 and they were on $100,000, and it took a loophole for that to surface. I had not noticed. Nobody had made me notice. If I had been paying the attention I have spent this chapter telling you to expect from the person above you, there would have been nothing to leverage, because the gap wouldn't have been there.

So don't read that as a warning about getting caught. Read it as the reason this chapter is about evidence and not about leverage.

**Evidence makes me pay you. Leverage makes me replace you.**

### The two conversations

With a reasonable manager. Ask for twenty minutes with a stated purpose, so nobody is ambushed.

"I'd like to talk about what I'm doing now versus what I was hired for, and where that should go next."

Then the case. What you took on. What it produced, in their units. What you want to own next. What the role should now be called. What the number should be. Then stop talking.

With a difficult one. The same case, plus one addition, because a difficult manager needs the decision made outside the room.

"I'm not asking you to decide today. I'd like you to have this, so that when the budget conversation happens you've got what you need to make the case."

That turns you from somebody making a demand into somebody handing over ammunition. It also survives a no, because a no becomes a not-yet with a mechanism attached.

If the answer is still no, ask one question and write down the answer. "What would have to be true for this to be a yes?"

A manager who can't answer that has told you something important about where you work.

### What you have really been doing

Everything in this book has been building towards something small.

You are making it obvious, to somebody who has to justify it to other people, that the business is better with you in it. The habits, the brief, the stack, the ten days all exist to produce one thing.

Something you can put on a desk in front of somebody who has to explain it to four other people.

So go and produce it. Then create the moment, and ask properly.

You already know things about your organisation that are written down nowhere. That was always the rarer half of it, and you had that half before you read a word of mine.

The other half was a method. You have it now.

Somebody in your building is going to become the person everybody comes to about this work. There is no queue for that, no application form, and nobody is going to nominate you.

Fifteen minutes tomorrow morning. One brief, written properly, for one task you already hate.

That's the entry fee, and it's the last thing this book will ask you for.

Go and write it.

---

One more thing, and it's the paragraph I started with.

**You are more powerful today than you have ever been in your life, if you can learn to drive this thing.**

It comes down to the way you think, the questions you ask, and how you assess what you're given back. You control an extraordinary asset now. Learn how, and you'll be paid more, you'll be worth more, and your future stops being something that happens to you.

When I wrote that on the first page, you had no reason to believe it. You'd read nothing, tried nothing, caught nothing.

Read it again now.

Every clause in it has a chapter behind it. The way you think. The questions you ask. The assessment. That is not a slogan, it is the table of contents, and you have just been through all three.

Which leaves the last clause, and it is the only one I could never write for you.

*Your future stops being something that happens to you.*

That one is not a technique and it does not arrive in an afternoon. It arrives the first time you catch something before it goes out and understand that nobody would ever have known if you hadn't. It arrives again the morning somebody two floors away asks for you by name — about work you were never assigned. And it arrives properly, one day, in a conversation about money that takes four minutes because everyone in the room already knows the answer.

I told you at the very beginning that somebody one desk over was going to learn this and you were not, and that neither of you would be told which one it was until the restructure.

That was true when I wrote it.

**It is not true any more, and you are the reason it changed.**

Nobody is going to hand you a certificate for that. There is no announcement, no moment where the building is told. What there is instead is a slow, unmistakable shift in who gets asked, and a version of you six months from now who would find this book slightly obvious.

Go and be the person they come to.