Education · December 2024
Educating together
First draft
Most of what I believe about education comes down to one bet: the highest-leverage thing we can do is make it easy for parents to learn alongside their kids, not to hand learning off to school and hope it comes back done.
Why "leave it to school" is the wrong default
School is the default container for learning, for a lot of good reasons. But it quietly trains everyone, parents included, to treat learning as something that happens somewhere else, on someone else's clock, in someone else's building. The kid goes, the kid comes back, and the day gets graded by whether the homework is done.
I don't think learning is a thing you outsource. The strongest signal a kid gets isn't the curriculum; it's whether the adults around them are still learning anything. Mine watch me far more than they listen to me. If I want curious kids, the cheapest intervention available is to be visibly curious in front of them.
"Leave it to school" also spends the best hours badly. The time a kid has with a parent is high-trust and low-stakes, the exact conditions the research says you want for learning. We mostly spend it on logistics.
The loop
I think of it as a loop with four parts, each one making the others better.
Kids bring the raw material: curiosity, and no fear yet of looking stupid. That's the asset you're trying not to break. Parents bring continuity and trust. We're the one part of the system that sees the whole child across years, not one subject for one term; we can't teach everything, but we can keep the thread. Schools bring structure and peers, the scaffolding and the social context. Most real learning is social, and you learn faster sitting next to someone else who's also trying. Communities bring range, the things no single family has: a neighbour who fixes engines, a library, a chess club, someone who has actually done the thing your kid suddenly cares about.
None of these works well alone. School without parents is a handoff. Parents without community is a bubble. The leverage is in the loop, not in any single node, and most of what's broken is simply that the parts barely talk to each other.
Learning intent is the real asset
A while back I sketched an idea I called a "learning counselor": something you could talk to in whatever channel you already use (a chat, a voice, a phone number) that asks one thing, what do you want to learn, and hands back a path. I was sure the clever part was the path it gave back.
I was wrong. The clever part is the question. Even if the counselor hands back mediocre content, you still walk away with a map of what people are actually trying to learn, captured honestly, before it gets bent to fit whatever courses happen to exist. That intent is the rarest input in education: we have endless content and almost no idea what any specific learner is reaching for on a given Tuesday.
It's the same at home. The highest-value thing I do as a parent isn't answering my son's question. It's noticing which questions he keeps coming back to. Intent first, content second.
And meet them where they already are. Don't make a kid, or a parent, come to a new app, a new place, a new ritual. Learning that needs setup mostly doesn't happen.
Where AI actually helps
This is where AI earns its place, and I say that as someone otherwise sceptical of dropping it into everything. Cheap, capable intelligence is starting to feel like a utility you can pull on tap, and this loop is one of the better things to point it at.
It changes three things. First, it removes the expertise gap. A parent doesn't need to know Python, or fractions, or the French Revolution to learn it alongside their kid; a good model is a patient tutor that explains in plain language and never tires of the fourth "but why". That frees the parent to do the part only a parent can do: keep the thread, stay curious, care that the kid gets somewhere.
Second, it is always on, and it meets you where you are. The best tutor in the world is useless at 8pm on a school night if it needs booking, setup, or a commute. An AI tutor lives in the tab, the chat, the voice you already use. Dropping the cost of setup to zero is the cheapest way I know to make the learning actually happen.
Third, it can hold the intent. Let a model ask what someone wants to learn, and listen properly, and you get that map I described, per kid, updated as they change. My son's curiosity in March is not his curiosity in June. A patient agent that keeps asking and remembers tracks that better than I can.
What AI does not do is replace the human parts. It is not the relationship, the motivation, or the trust; those stay ours. Used well, it removes the friction (expertise, availability, setup) so the human parts have room to work. Used badly, it becomes one more thing we outsource the learning to, which is the exact mistake I am trying to avoid. The tutor sits alongside. It does not take over.
Why I started with code
All of this is easy to say and hard to build, so I started with the smallest concrete thing I could actually finish: a coding tutor for my two boys.
It's a browser-based classroom. Each kid opens a URL and gets their own sandbox and a terminal right in the tab, no install, no IDE, nothing to set up. A day-by-day curriculum, and an AI tutor alongside that talks in plain language: brief, simple, a little fun, the way Feynman would. (Getting the thing to stay running was its own saga. Different post.)
I picked code for selfish reasons. It's the subject I can teach from first principles, and the feedback loop is instant: you run it, it works or it doesn't, and the kid learns to debug instead of to wait for a grade. But the point was never to turn them into programmers. It was to sit next to them while they learned something, and to find out whether the loop I keep describing actually closes when you build a piece of it.
It's an early attempt, and I have more questions than answers. How much structure before it stops being play. How to pull in the community part without it becoming another logistics chore. Whether any of this survives contact with two tired kids on a Wednesday. But the bet underneath hasn't moved: don't outsource the learning. Do it together, and build the tools that make doing it together easy.