The Contract

Part I was the diagnosis. Some of it probably landed harder than you expected . The FONA is real . THE NEW TIRED is real. The codebase you built and are now afraid of: apparently also universal , since that's what opened this chapter.

So now what .

The honest answer is that most of this doesn't have a clean fix. The technology is not going away. The pressure is not going away. The implicit SLA, the always-available floor, the comparison to a thing that doesn't have bad weeks: also not going away . You don't get to renegotiate those . They are the new conditions .

What you can renegotiate is the internal contract. The private arrangement you have with yourself about when to ask, what you trust , what you own, and what counts as yours when you're done. Most people don't have one. They have habits. Some of those habits are bleeding them quietly.

This is not twelve steps to mindful AI use, which is a phrase that should never appear in a book and I apologize for almost writing it. It is a set of honest moves for a Tuesday.

Not in order. Pick the one that fits the problem you're having right now, which is probably one of three: you're prompting anxiously, you're running on empty, or you've built something you can't read and are slowly becoming a prompt jockey for your own project .

All three are addressable . None are fully fixable . Welcome to Part II.

***

What is the contract actually?

The old one was simple, even if nobody wrote it down. You did the thinking . You did the typing . You owned what came out . Effort and authorship traveled together because they had to. There was no other way to make a thing. The work you put in was the proof the thing was yours .

The new contract snuck in. Nobody signed it. It just happened , one prompt at a time. The model does some of the thinking. You do less of the typing. You still own what comes out, your name is on the commit, your byline is on the report, your address is in the From field. But there's a gap now between what you own and what you understand. You can sign things you couldn't build from memory. You can ship things you couldn't debug at 3 AM if they broke.

That gap is the problem. Not the model. Not the tool. The gap.

Authorship without understanding is fraud. Refusing the tool entirely is martyrdom, and the world has stopped clapping for martyrs. Neither works. The renegotiation is the middle: use the tool , stay the thinker. Outsource the typing , keep the understanding. Let the model write the first draft. Just know what's in it before you put your name on it.

This is not a how-to. It's a set of small refusals and small actions that hold the line. Six of them. Pick the ones that fit. Skip the ones that don't.

***

Ask the question in your head first.

Before you type the prompt, ask the question in your own head , in one sentence . Out loud if you have to , weird neighbors be damned . If you can't phrase it cleanly , the problem isn't ready to be asked . You're not looking for an answer yet. You're looking for what's actually bothering you , which is the real work, and the model can't do it for you.

This is the best filter you have. The clearer you can ask the question , the less you need the model to answer it . Half the time the answer is sitting on the back of the question , and asking it well surfaces both. The model just confirms what you already knew. The other half, the question is genuinely hard, and now you have a sharp prompt instead of a fuzzy one , which is the difference between a useful answer and a paragraph that sounds smart and means nothing .

If you can't say the question , don't type the prompt. Sit with it for thirty seconds. The asking is the thinking. Skipping the asking is the model doing the work while you sit in the audience.

***

Say it back.

After the model gives you something, code, a draft, an analysis , an answer , explain it back to yourself in your own words . Without looking at the output . Out loud or on paper . If you can, the thing is yours. You can defend it , modify it, ship it.

If you can't , you don't own it . You are hosting it . It is passing through you with your name attached and you do not know what's in there. This is the moment the gap widens . The fix is not glamorous: read it again , ask the model to explain the parts that didn't track, write the explanation in your own words until it sticks. Do not move on until you can say it back without notes.

This is tedious. It is also the difference between knowing what you built and being the courier for what the model built. There is no shortcut. The shortcut is what got you to the chapter you're afraid of.

***

One file at a time.

For the app you're afraid of. The codebase. The doc you've been generating into for three weeks. Whatever it is that has gotten too big for you to hold in your head .

Before you touch any part of it , read the part you're about to touch. One file. One function. One section. Not the whole thing , you don't have time for the whole thing , you didn't have time for it three weeks ago which is why you're here . Just the piece relevant to the change . Read it. Understand it. Then prompt.

This is slower than not doing it . Yes . That's the point . Speed is what got you here. Every change you make to code you've read is owned. Every change you make to code you haven't is borrowed , and borrowing compounds. The dread in the back of your mind is the compound interest. The only way to pay it down is to read.

You will not catch up in a weekend . You don't need to . You just have to stop adding to the debt. Read what you touch. Touch one thing at a time . The debt stops growing the moment you do this. That alone changes how the project feels in your hands.

***

Did I want this before you offered it.

The model will offer you things. Features it could add. Refactors it could do . Sections it could expand . Improvements that sound completely reasonable and have nothing to do with what you came here to build . For each one , ask: did I want this before you mentioned it.

If no, no. Not because the suggestion is bad. Because it's drift , and drift is how the project you started becomes the project you can't remember starting. The model is calibrated to what sounds plausible next, not to what you were trying to do. It doesn't know what you came in here for. You do. Or you did, an hour ago, three suggestions ago , before the conversation took a turn neither of you remembers .

The question has to be reflexive . Did I want this . Did I come here for this . Anything other than yes is a no . You can always come back and add it later , if it turns out to be a real idea and not just a thing that sounded plausible in the middle of an unrelated conversation.

***

The unbuilt list.

You have a list. Mental or actual, doesn't matter. The newsletter you keep meaning to start. The side project from 2021. The app idea you've sketched three times on three different napkins. The novel.

Stop carrying them .

Pick one. Decide. Build it this month, or release it. Not "release it for now. " Release it . Cross it off . It was not a project . It was a feeling that wore a project costume.

The carrying is the tax. FONA collects it every day you keep these ideas alive without acting on them. They generate guilt at a steady rate . They produce nothing. They sit in your head like browser tabs you will never read , except they have your name on them , so you can't close them without admitting something.

Admit it . Decide . One a week if you have to . The relief of cutting an idea you've been carrying for three years is the closest thing to free time the renegotiation hands you.

What you keep , build. What you cut , let go for real . Not for now. For real.

***

The unprompted hour.

One hour a day, don't ask the model anything. Not a detox. A calibration.

You need to know what you can still do alone. If you can't write the email , debug the function , draft the message, decide the thing , without the tool: you need to know that. Not as a failure. As information . The gap you can't see is the gap you can't close , and the only way to see it is to work without the tool for a measurable stretch and notice what happens.

Most people , when they try this , are surprised twice . First by how reflexive the reach for the model has become. The hand opens the chat before the brain decides . Second by how much they can still do . The atrophy is usually less than the dread suggests. You are still in there. The hour is how you find out .

Pick the hour. Same time every day if you can manage it, whenever if you can't. Skip days. Come back. You will know it's working because the gap will feel smaller , which is the only metric this chapter has.

***

Back to the app.

You built something you're afraid of. The gap between what you own and what you understand is real . It is wider than it was last month. You have been managing this by not looking directly at it.

You can close the gap. Not all at once. Not by stopping and starting over , which is the move FONA will keep offering you , because starting over feels like a clean slate and a clean slate feels like control. It isn't. Starting over is just opening a new tab to be afraid of in six weeks.

The way out is smaller. Hold the line where you are. Then walk it back . The next prompt: ask the question in your head first . The next change: read the file . The next suggestion from the model: did I want this before you offered it.

The contract is small. It is not heroic. It does not feel like much when you are doing it . The doing of it is the renegotiation. That's the contract .

The tool was never going to give you back your day. You have to take it.

***