What we actually mean by supervision
Ratios, sightlines, chat logging, screen visibility. The unglamorous mechanics of how a room full of kids and AI tools stays supervised, hour to hour.
October 2, 2026 · Hi, Bot
"Supervised" is a word that shows up on every parenting form and means almost nothing on its own — it's the kind of claim any program can make without saying anything specific. We'd rather be specific. Here's what supervision actually looks like in our room, mechanically, not as a marketing sentence but as a set of decisions we made on purpose and can defend one at a time.
Ratios, and why we don't flex them
We run a fixed adult-to-kid ratio, and it doesn't loosen on a busy day. A mentor watching too many kids stops being able to tell the difference between a kid who's productively stuck and a kid who's actually struggling — that distinction is the core of the job, and it degrades fast past a certain group size. When a session would push a mentor over ratio, we cap enrollment for that session instead of stretching the adult thinner. That's a real cost to us in seats sold. We hold the line anyway, because the ratio isn't a nice-to-have, it's the thing that makes the rest of supervision possible.
Screens face out, not in
Every workstation in the room is arranged so a mentor walking the floor can see the screen, not just the back of a kid's head. This isn't about distrust of any specific kid — it's a room-design decision that removes the need to guess. A mentor doesn't have to interrupt a kid to check what's on their screen; they can see it from six feet away while walking past. That single layout choice does more supervisory work than any policy we could write down.
What gets logged, and what doesn't
Conversations a kid has with an AI tool in our room are logged — not surveilled in real time by a human reading every line, but retained and reviewable if something looks off, and reviewable by a parent on request. We tell kids this up front, plainly, not as a gotcha. It changes behavior in a good way: knowing a conversation could be looked at later is itself a mild, healthy check on both the kid and the tool. We don't log anything happening on paper, in a notebook, or in a conversation between two kids at a table — supervision covers the AI interaction specifically, not every private thought a kid has in the room.
Age-appropriate access is enforced at the account level, not the honor system
Which models and tools a kid can reach depends on their age band, and that's enforced by what their account can actually open, not by asking a ten-year-old to self-regulate. We've written separately about why the age line moves the way it does; this post is just the mechanics — the account itself won't render a tool that's out of band for that kid, so there's nothing to police in the moment.
What supervision doesn't mean here
It doesn't mean a mentor standing over a kid's shoulder narrating every keystroke — that's not supervision, that's interference, and it kills the exact kind of independent problem-solving we're trying to build. It also doesn't mean a kid is ever in a room alone with a screen and no adult present; there's always a mentor on the floor, visually scanning, even during the stretches where nobody needs them. Supervision here is closer to a lifeguard's job than a teacher's: mostly watching, rarely intervening, always positioned to see the whole pool.
Why we're this specific about it
Vague reassurance is cheap and parents have learned, correctly, not to trust it. Ratios, sightlines, logging, and account-level gating are all things you can ask us about directly and get a real answer to, because they're decisions, not slogans. If any of them ever changes, it'll be because we found a better way to do the same job — not because it got quietly loosened for convenience.
› Keep reading
Get the next dispatch.
Occasional, never spammy.