Every school has an AI policy. Every kid has an AI workflow. They are not the same thing.
I asked all three of my kids the same question separately: what is your school’s rule about AI?
Then I asked a slightly different question: what is the actual rule among students?
The second question was much easier.
Violet’s version was essentially that everyone assumes people use it, you don’t get caught, and you don’t advertise what you’re doing. Warren said almost every student uses AI for something: writing, checking writing, research, even mindless work like moving information into a lab spreadsheet.
Jasper gave me the most efficient answer. I asked what the real student rule was.
“Using AI.”
Two words. That’s it.
His school handbook, as he described it, was more nuanced: don’t use AI if you think your teacher wouldn’t allow it. Which is less a policy than a vibe.
This is not me attacking schools. Teachers have been handed an ugly problem, and generative AI went from novelty to normal student workflow faster than most institutions can approve a new lunch vendor. But the gap between the official workflow and the real workflow matters, because the kids aren’t just learning how to use AI. They’re learning how to manage the adults around their use of AI.
A rule everybody routes around teaches something
Violet’s false accusation, from the last installment, was one cost of that gap. After interviewing all three kids, I saw another.
A rule that sophisticated students can route around and honest students fear doesn’t teach integrity. It teaches concealment.
Warren was blunt about this: it isn’t hard to avoid detection if you aren’t stupid about it. Which creates a strange incentive ladder. The student who pastes an entire AI-generated essay into a document without reading it may get caught; the student who uses AI carefully, iteratively and intelligently probably won’t; and the student who doesn’t use AI at all may still get accused.
I’m not sure that is the behavioral ladder we intended to build.
The OECD’s 2026 Digital Education Outlook draws the line schools actually need here. AI can improve a student’s output without improving the student. Used as part of good teaching, it helps learning. Used to outsource the thinking, it makes the artifact better and leaves the human roughly where they started.
That is a far more useful distinction than “used AI” versus “didn’t use AI.”
“Used AI” describes almost nothing
There is a difference between “write my paper” and “here is my argument; tell me where it is weak, but don’t rewrite it.” There is a difference between having AI solve the math problem and asking it to explain why your answer is wrong. One outsources the thinking. The other uses a tool inside the thinking.
The kids understand these distinctions because they live inside them. Policies often don’t.
We regulate the noun because the verbs are harder.
Our summer experiment made that impossible for me to ignore. I wasn’t trying to teach the kids how to get ChatGPT to write school assignments. I was trying to see whether relatively inexperienced people, handed a structured environment of tools, methods, feedback loops and guardrails, could do sophisticated work safely.
The work did not divide neatly into “AI did it” and “the kid did it.” It was a loop: the kid decided what to build, described it, got output, judged it, corrected it, got stuck, changed direction, and occasionally asked why the system had apparently decided to ruin their afternoon.
That is increasingly how adults work too.
Pretending students inhabit a different technological universe doesn’t prepare them for ours.
Disclosure only works if honesty is safe
The obvious answer is to require students to disclose AI use. Directionally, I agree. But disclosure only works if telling the truth doesn’t automatically incriminate you.
Imagine telling employees, “You may use Excel, but every time you use a formula, please disclose it so management can decide whether you really did the math.” People would stop disclosing formulas.
If a student believes saying “I used AI to critique my outline” creates a meaningful chance the teacher will treat the paper as suspect, the student learns not to disclose the AI. That isn’t a character flaw. It’s incentive design.
A better disclosure is more specific. What did you ask AI to do? What did you not allow it to do? What did it get wrong? What did you change? Which part of the final work are you prepared to defend as yours?
Now disclosure is part of the learning rather than a confession.
Bring the real workflow into the room
The answer cannot be “anything goes.” A student who asks AI to produce an essay they never read has learned very little. A student who cannot explain the work does not own the work.
But if students are already using AI, the opportunity is enormous. Have them compare three AI answers and identify the strongest. Give them a confident hallucination and make them find it. Let them use AI on an essay, but require a visible revision trail and a short oral defense. Ask them which parts of the project should not be delegated. Have them document where the model failed and what they did about it.
Grade the judgment and the reasoning, not the shininess of the artifact.
Those are not concessions to AI. They are increasingly the skills required to use AI without becoming an idiot with a very productive assistant.
UNESCO’s guidance lands in roughly the same place: keep human agency and critical thinking at the center. The goal isn’t maximum AI. It is better humans using AI on purpose.
The curriculum nobody wrote
What worries me isn’t that teenagers are using AI. Of course they are. Adults are too.
What worries me is that we may accidentally create two curricula. The official curriculum teaches students to produce work as though AI doesn’t exist. The unofficial curriculum teaches them how to use AI without adults noticing.
Guess which one teenagers will get better at.
A rule everybody quietly violates does not teach integrity. It teaches that rules are theater and adults can be managed. That may be the worst AI lesson we could possibly give them.
The alternative isn’t surrender. It’s honesty. Bring the real workflow into the room, then teach students how to use it well enough that they don’t have to hide it.
Next: I asked all three kids whether they thought they would need to write well as adults. They unanimously informed an author that writing is basically over. This went well for me.
Sources and further reading
Want this energy at your event?