What You Certify

You have been thinking small .

That is not an insult . The first seven chapters of this book have been about you and your team, your habits and your norms, and that is the right scope for figuring out how to use AI without lying to yourself in a Tuesday meeting . The scope has been useful.

It is also a fiction.

Look. You are not just using AI. Your whole field is being rewritten around it , and you do not know any of the people doing the rewriting.

Somewhere there is a small group of them . A licensing board. An accreditation committee. A regulator three levels deep in some agency you have never heard of . A standards committee where seven people are arguing over a footnote . They are deciding what counts as

legitimate work. What you have to disclose. What your license still means . They are not asking you . You will not get a memo.

None of this is news. Each is a small adjustment to the rules of your work , made by the small number of people who get to make those rules.

In five years the profession you are in will be different from the one you trained for . Not because the technology changed. The technology changed in 2022. The profession changes in 2026 and 2027 and 2028 , in rooms you have never been in.

You will find out the way you find out about every other professional rule change. A new module appears in your annual training. A new line shows up on your malpractice form . A junior at your org mentions something has shifted. A certification you held in 2021 gets labeled legacy. You will adapt. You will adapt slightly behind whoever was paying attention .

This chapter is for the person who would rather pay attention.

***

Watch the rooms you are not in.

There is a PDF that has already been written about your job. You have not read it . It is online. It is free.

It is the position paper your industry's standards body put out six months ago about AI use in your field. Or the consultation document the regulator published last quarter. Or the minutes from the licensing board's January meeting where they debated, then voted on, whether AI-assisted work counts toward your annual training. The document exists. It is searchable. It will shape your work in the next two years.

You have not read it because nobody you work with has read it. Reading these documents is not part of your job description , was not on your manager's last 1:1 agenda , and does not show up in any performance review. It is also where your profession is being rewritten, in detail, in language you can mostly understand if you sit down with it for an hour.

The Move 1 : pick three.

Three professional bodies , regulators , or standards committees that govern some part of your work . Find their website. Subscribe to their newsletter. Read one document a month. That is all. One hour a month, total.

You will learn things nobody else on your team knows . The licensing body's AI guidelines . The regulator's draft rule. The trade association's position . What was proposed. What got pushed back on. Who pushed back. What is coming next. This is not insider information. It is just unread.

A year from now , when one of these decisions lands on your desk as a new rule , you will already know what it is, why it was made, and what the alternatives were. Everyone else will be reading the email for the first time. The gap between you and them is the gap between knowing and finding out late.

That is the smallest move in this chapter . It is also the move that makes every other move possible . You cannot show up to the rebuild if you do not know where the rebuild is happening.

***

Disclose more than you have to.

You have been quiet about it.

The report you sent the client last quarter. The brief you filed. The strategy memo. The architectural decision record. The investor update. AI helped you produce each of them. You did not mention it. You did not have to. The recipient probably assumed , or did not think to ask , or did not want to know.

Everyone you work with is also quiet about it. Across your industry, professionals are using AI in their work and not flagging it in the work itself. There is no rule yet. The rule is coming. The question is what shape it takes when it arrives.

The Move 2: disclose AI use before anyone makes you. Not all of it. Not in a way that turns every report into a confession. Specifically and clearly when it matters: when the work product is published , when the deliverable is high stakes , when somebody downstream is going to make a decision based on it.

A line at the bottom of the report. A note in the methodology section . A sentence in the cover email . Portions of this analysis were drafted with AI assistance and reviewed by [your name] . Six words plus your name. That is the move.

Two things happen when you do this.

The first is that the people receiving your work get used to seeing AI flagged, before the rule lands . They learn what acceptable disclosure looks like because you showed them. The professionals who normalize disclosure in their field shape what disclosure means when the regulator finally writes the rule. The ones who do not get the rule that gets written without them.

The second is that you have made the disclosure on the record. When the rule arrives, when the lawsuit happens , when the regulatory inquiry comes, the work product itself shows you disclosed . You said what you used and what you reviewed, at the time. The people who said nothing are about to spend a year explaining why they said nothing.

There is a third thing , harder to measure. Disclosure changes how you do the work. The moment you know you will name the AI involvement publicly, you pay closer attention to what it is . The verification gets tighter . The review note from Chapter 8 becomes a habit , because you know you might have to defend what you signed.

You will resist this move. It will feel like overshare, like making yourself an obvious target, like inviting questions you would rather not field. Do it anyway. The professionals who set the disclosure norm are the ones who chose to disclose early. The ones who waited are the ones the rule was written for.

***

Build the audit trail before it is required.

You used to be able to save the conversation. Copy the chat. Paste it in a doc. Done. That was 2024.

The work has changed.

An agentic session runs for hours, spawns sub-agents, calls tools, edits files, produces outputs you never directly typed. A working engineer triggers 100 to 200 inferences a day. Manually logging any of it became fantasy somewhere around the third quarter of 2025.

Nobody is going to stop after every AI action and write a compliance note. They should not have to.

The Move 3: is not to ask humans to document the audit trail. The move is to capture it from the work already happening.

The task lives in the Jira ticket. The boundaries live in the SPEC.md, the ADR, the RFC. The files changed live in the commits. The tools that ran live in CI logs.

The human edits live in the PR. The approval lives in the merge. None of these are new artifacts. Your team already produces them. The audit trail is the chain that connects them, with AI involvement marked at each link.

A real audit trail looks like a flight recorder. Quiet while the work happens. Useful when something goes wrong. Clear enough to answer: what changed, who accepted it, what evidence existed at the time.

"We reviewed it" is going to stop being enough. The future audit question is not show me the chat. It is show me the decision path. Show me the ticket. Show me the SPEC.md document. Show me the commits. Show me where the human stamped this as done.

Regulated industries get there first. Finance, healthcare, law, defense, anything attached to compliance. Within two years they will need evidence of human oversight on demand. Not vibes. Not memory. Evidence. Outside those fields, the first major incident where AI-assisted work causes harm and nobody can prove what was reviewed will reset expectations across the rest.

The other reason to build this is not compliance. It is learning. A trail shows where the agent helped, where it drifted, where humans corrected it, where the same mistake happened three times. That is operational learning, not bureaucracy.

The audit trail of the AI era is not a transcript. It is a system of record for human judgment.

Build it before someone asks you to prove it.

***

Use your tokens like they are yours.

You are using AI like it is free.

It is not free. It is being subsidized. Venture capital is paying the difference between what your prompts cost to run and what your company pays for them. The subsidy will end. Pricing models will tighten. Token budgets will arrive. The free era of any-prompt-any-time is the version of this technology you happen to be using in 2026, not the version you will be using in 2028.

Most professionals are spending tokens like the bill is somebody else's problem. The bill is going to land. When it does, the people who already know how they use AI will absorb it. The people who built the reflex of throwing every problem at the model will discover that half of those problems were not worth a prompt.

The Move 4: spend tokens deliberately.

For one week, before every significant prompt, ask one question. Is this worth the spend. Not is the model capable. Not will it save me time. Worth the spend, where the spend is real cost to your company, finite capacity for your team, and time you will not get back if the output is bad.

Most of your prompts will pass this test. Some will not. The ones that fail are the prompts that were reaching for the model because it was there, because it was free, because the alternative was sitting with the problem for ninety seconds.

The professionals who survive the metering era will be the ones who built this reflex before the meter arrived. The professionals who did not will have to learn it the hard way, with their company breathing down their neck about quarterly AI spend, in a conversation that does not consult them about the previous two years of habits.

You can start the reflex today. The meter is coming whether you start or not. The cheapest version of starting is now, while there is no penalty for the prompts you waste.

***

Some signals are gone.

The last time you really knew what one of your teammates thought, really knew (from how they put it together), was probably eighteen months ago.

This is not nostalgia. It is a thing that used to work that has stopped working.

Slack messages used to have personality. The clipped reply. The slightly-too-long explanation. The typo nobody fixed.

The sentence that ran on because the person was excited. You knew who was annoyed, who was confused, who was actually thinking it through. You read your team the way you read friends. Not perfectly, but well enough.

That signal is gone.

Every Slack message is articulate now. Every PR description has the same three bullets. Every standup update is appropriately confident. The doc your direct report sent reads like the doc your other direct report sent. The reports are not the same person. The output is.

Writing was hard. Difficulty is what carried personality through. Now writing is easy. Personality has nowhere to ride.

You will read arguments online about whether this or that signal can be saved by tightening the rules. Most are written by people who built careers reading those signals. Hear them out. Then notice that nobody talks about typing speed at job interviews anymore either, and the people who used to read it found other things to read.

The Move 5: stop pretending the dead signals are alive.

When you read a Slack message, do not weigh the tone. When you read a PR description, do not assume the voice is theirs.

When you read a self-assessment, do not believe the cadence tells you who wrote it. These artifacts still happen. They just stopped carrying what they used to carry. Use them to start the conversation, not finish it.

This is the easiest move to describe in this chapter and the hardest to do. You have been calibrated to these signals your whole career. Trusting your read of a Slack message was a skill. The skill is no longer useful. Setting it down is grief, even if the loss is small.

You will need the attention you free up. The new signals are still being built. The faster you stop spending energy on what is gone, the more you have for what is being built.

Some things are not coming back. Move on. The next thing needs you.

***

Choose what you certify.

You signed something today. Maybe several things.

A PR you approved. A doc you sent. A report you submitted. A reply you wrote. A merge you pushed. A chart you presented. A decision you made on behalf of a customer. Each one had your name on it, in some form. Each one was an act of certification. You said: this is mine, I stand behind it, I will be the one if this is questioned later.

"How many of those things would you describe as fully yours.?"

This is not a guilt question. The honest answer for most readers, on most days, in 2026, is some of them. Not all of them. The model was involved in most of them. Sometimes you reviewed it carefully. Sometimes you reviewed it less carefully. Sometimes you skimmed and trusted that nothing important was in there. Sometimes you knew you should have read it and you sent it anyway because the meeting was in seven minutes.

This is the situation. The book has been about this situation. Every move in every chapter has been one way to close the gap between what you sign and what you stand behind.

Maybe you chose this technology. Maybe somebody chose it for you. Either way, it is here. In your industry, your company, your tools, your career, your reading list, your annual training, your performance review, your hiring loop, and now your professional licensing body. Some of you ran toward it. Some of you braced against it. The result is the same. You work with it now. The choice in front of you is not whether to use it. The vote was held without you. The result was unanimous because nobody asked the question out loud.

The Move 6: is you choose what you certify.

That is the only thing the model cannot take from you. Your signature is a claim: a human looked at this, understood it, and will be the one accountable if it goes wrong. The technology can write the work. It cannot make the claim. Only you can.

For everything that ships under your name, ask one question. Did I do enough that this is mine. Not did the work get done. Not did the deliverable look good. Did I do enough that, six months from now, in front of a regulator, a customer, a court, a code review, a peer, or my own kid asking what I do for a living, I am willing to point at this and say: I made this.

If the answer is yes, sign it. Stand behind it.

If the answer is no, do more work, or take your name off it, or send it back to whoever is asking you to sign for less than the work you actually did. The third option is harder than it sounds. It will be the move that defines your career in this decade, more than anything else in this book.

You did not get to choose the technology. We get to choose what we let it certify.

The book is done. What you sign next is what the book was about.