I'm Arming My Sons With AI to Win at School

Our sons go back to school next week, one into eleventh grade and one into his sophomore year of college. Before they do, we're setting each of them up with a system that gives them a study assistant for every class they're in.

We want two things out of it. First, we want school to be as low-stress and as mentally healthy for them as it can be. Second, we want the maximum return on the hours they're already putting in.

We ran a small version of this at the end of last year, to help study for finals. It worked. Anxiety went down and grades went up, in that order, and the two seem correlated - though two kids and one set of finals doesn't prove anything, and I won't pretend otherwise. What we're building now is substantially more than that. I'm writing about it before the results are in because waiting until June to report back won't help anyone headed to school next week.

I'm optimistic. I've built a lot of systems like this, at work and at home, for my own writing and my own projects, and they've worked. They also get better every month. The models keep getting bigger, and everybody using them is working out better techniques and better tradecraft all the time.

What follows are chronic gaps in how school works. None of them are new. What's new is that there's finally something you can do about them.

Memory and Notes

What a student gets out of a class is capped by how well he understood it in the moment and how much of it he wrote down. Neither is a hundred percent. He got distracted, or didn't follow it while it was being said, or was busy writing one thing and missed the next. Note-taking is its own trap: writing down what was just said takes the attention he needs for what's coming next.

For as long as there have been classrooms, the fix has been to tell the kid to pay closer attention and take better notes.

An AI-accessible transcript gives him the complete record, and then he can work with it. Ask what was covered while he was distracted. Find where a term was first explained three weeks ago, in his teacher's words rather than a textbook's.

It also means he doesn't have to write anything down to begin with, so he can listen. And the record is what makes the rest of this possible. With the lesson written down, it can come back at whatever speed and in whatever form works for him.

One caveat. Writing is itself a way of learning, and for plenty of people the act of writing is how the material goes in. Keep taking notes. The transcript is extra.

Pace

Kids move at different speeds. Some need to marinate with an idea, poke at it, take longer with it. Others pick it up fast, depending on the day and the subject and a dozen other things.

The class moves at one speed anyway, because it has to. There's a curriculum. There's a fixed number of weeks. There's standardized testing at the end, which means a certain amount of material has to get covered in checklist fashion. Schools do build in extra time in various ways, but inside the primary instruction period, it goes at the speed it goes.

That's structural, not a teacher problem. There are teacher problems. This isn't one of them.

A recording has no speed at all. Once the lecture exists as text, it can be revisited, chunked up, paused, sped through on the parts already known, slowed down on the two minutes that didn't land. The class period doesn't go away - the kid still sits through the same fifty minutes with everyone else. Those fifty minutes just stop being the only shot at it.

And this runs in both directions. Differentiated learning has been claimed almost entirely by the intervention side, understood as something for kids who are behind. It's equally for the kid who had it in the first four minutes and now has twenty minutes of nothing. Dead time is its own problem and it's the mirror image of needing more time, and schools address it hardly at all. That time could go to reflection, or a break, or something else productive. The idea is not being hostage to one timeline.

I'm not claiming a transcript magically corrects any of this. But there's something there.

Modality

Some people learn by reading. Some by listening. Some are multimodal, and if you're dyslexic in particular, a combination of listening and looking and writing and talking is often what gets a thing to stick. Some people need an analogy. Some need it chunked. Some need plainer language, and some need it sped up because the pace is putting them to sleep.

In school, if it's a worksheet, it's a worksheet. If it's a video, it's a video. That format wasn't chosen for any particular student. Nobody could have run thirty versions of the same lesson, so one version got made.

I don't think there's a single kid in that classroom for whom the material is pitched perfectly. Maybe there's one. Not the majority.

Once the class exists as text, the format is no longer fixed. The same lesson can come back as a conversation, a diagram, a set of practice questions, an analogy said out loud, whatever gets it to stick. And a kid can go looking on their own - ask the model to try four different ways of explaining it and see which one works for them. That's a different thing from being assigned a format. I don't know all the ways people learn, and there are probably dozens, and you don't need to know them either to start trying them.

Pooled Notes

A transcript gets you what was said in the room. It leaves you stuck with your own understanding of it.

You still take the class in through your own head - your hardware, your software, your lived experience, the connections you happened to make. Read the transcript again and you're still bound by the same limits. Meanwhile there are thirty other people in that room running the same material through completely different equipment. What they took away, where they got stuck, how they'd explain it to somebody else, all of that is interesting, and all of it is locked in their notebooks.

People already do a version of this. They study together, they swap notes. It's high overhead, it's bilateral, it's usually one person's notes at a time, and it's a favor you have to ask for.

Now imagine everybody put their notes into a pool, and that pool became another input alongside your own notes and the transcript. What shows up across all of them at once is more interesting than any single classmate's version. Where six people all struggled to articulate the same thing, that says something about the material. Where one classmate explained it more clearly than the teacher did, you can use their version. The spots where people disagree, or can't quite put it into words, are the spots to go study.

Psychological Safety

A student has a handful of ways to get an individual need met in a classroom. Answer a question posed to the group. Raise your hand and ask for help. Say out loud that you didn't follow that last part, or that you missed it entirely. Every one of them spends the room's time. Thirty other kids stop listening to the teacher and look at you instead, and the curriculum waits.

There are two costs in that, and both of them make a kid less likely to ask.

The first is efficiency. A question is an inefficient way to run a room unless everybody benefits from it, and the mileage on that varies enormously. Sometimes it's the question everyone had. Often it helps one person and does nothing for the other twenty-nine. Nobody can tell which in advance, so asking is a gamble with other people's time, and the student is the one placing the bet. A kid who doesn't ask isn't being irrational. He's reading the room correctly: thirty people, a checklist to get through.

The second is social, and it's the familiar one. How asking makes you feel about yourself. How it positions you with everyone else. And the one that lasts: anything you say in that room can follow you out of it. A question isn't a transaction that closes at the bell. It becomes a thing people know about you. That's a lasting cost for a momentary need, and deciding not to pay it is a reasonable call.

Both go away when a kid can get that individualized help without having to pay for it. There's no room's time to spend, no audience, no record anyone else reads, and nothing follows him into the hallway. He can ask the dumb question, ask it four more times, and say plainly that he's lost. That's true for the kid who was embarrassed and just as true for the kid who was being polite, or who didn't want to slow everyone down over something he figured he should already know.

It's the same structure behind the pace problem, seen from the other side. One kid's needs and the room's needs are in tension, and the room wins.

Patterns, Week by Week

School assesses in snapshots. Test, grade, move on. Nothing looks across the arc of how one kid's understanding actually developed.

And nobody should be waiting for the next report card to look back, because by then it's too late to do anything about it. Every week of saved class material says a little more about what's working and what isn't and why, while there's still time to change something.

The useless version of this is "you struggled in chapter four." What I want it to tell him is what kind of material gives him trouble. Abstract concepts. Anything delivered as a lecture instead of a worksheet. Anything the class moved through quickly. That's the difference between knowing where a grade dropped and knowing what to do about it, and grades on their own can't get you there. It works because the transcripts and handouts are sitting right next to the test scores.

One of our sons was out sick a lot one year, and re-entry was always hard. The trouble didn't show up the week he was out. It came later, on a delay, and not in the way you'd expect. That lag is exactly what a person misses, because by the time the problem appears the absence has scrolled off and nobody has both ends in front of them at once. A teacher with thirty kids and one year of them can't see it. Neither can a parent working off report cards.

Nobody has to turn any of that on. It falls out of keeping the class material in one place instead of throwing it out every week.

Quality Floor

Not every teacher is a ten out of ten. That isn't a controversial thing to say about any other profession and it shouldn't be here. What's different about this one is what happens next: a poor teacher is rarely held to account for it, and the stakes are a year of somebody's kid.

A model isn't better than a great teacher. It's reliably better than a bad one, and it never has an off year. That's a floor under teaching quality rather than a ceiling on it. Which teacher a kid draws in September is luck, and no parent gets to fix that. A study assistant in every class raises the worst case.

Back to School

Those are the gaps. None of them are new, and until recently none of them were fixable at the level of one kid.

Our boys go back next week, and each of them is going in with a study assistant for every class, set up against every item on that list.

If you want to know how it's built, comment, like, or repost. If enough people do, I'll write that one up.

(Curious how this article was developed in partnership with AI? Some of it is on my new YouTube channel, Both Hands on the Wheel.)