What we mean when we say "build with AI"

We say it on the homepage. We say it in the mentor onboarding. We say it to parents on the tour. Kids who can build with AI. Three words doing a lot of work.

Here is what we mean by them. Specifically. With receipts.

Build, not use

There is a very large and very loud category of programs right now teaching kids to use AI. Type a prompt, get a paragraph, paste it into your homework. The room is full of laptops and a worksheet that says Try these five prompts. The kid leaves having generated a thing she did not understand and could not reproduce.

That is not what we mean. Using AI is a verb that depreciates. The product changes; the buttons move; what you learned last year stops working. A kid who only learned to use AI is a kid whose skill is tied to a specific app's UI on a specific Tuesday.

Building with AI is different. Building means: the kid understands what an API is. Understands what a model returns and why. Understands that the thing in the browser is one of several front ends to the same underlying machine. Knows how to wire a model into something she made — a Pokédex for her sibling, a study buddy for her own SAT prep, an agent that plays a tournament for her. The artifact is hers. She could rebuild it on a different model. She could rebuild it next year.

That's the difference. One depreciates. The other compounds.

AI is a force multiplier, not a substitute

The pedagogical mistake we see everywhere — schools, camps, edtech apps — is treating AI as a replacement for the foundation. Why teach kids HTML when AI writes HTML? Why teach kids to read code when AI reads code?

Because a multiplier of zero is zero.

If a kid does not understand the request/response cycle, AI cannot help her debug a broken API call — she will not even know what's broken. If a kid does not understand what a function returns, AI cannot help her tune an agent — she will paste the model's output into the wrong slot and the agent will silently produce garbage. If a kid does not understand that data has a shape, AI's confident wrong answer will look the same to her as its confident right one.

So we teach the foundation. The request/response cycle. The shape of data. What a function is and what it returns. Enough of the actual machine that, when she sits down with an AI tool, she is directing it instead of guessing alongside it.

Then the multiplier is real. Then a twelve-year-old can ship in a week what used to take a college freshman a semester.

The verify habit

The other thing we teach, which almost no other program teaches and which we consider non-negotiable: don't ship code you can't verify.

AI generates plausible code. Plausible is not the same as correct. A model will confidently invent a function that doesn't exist, hallucinate a parameter, mis-spell a method, get the off-by-one wrong, swap two arguments. It will do this in the same calm voice it uses when it is right.

A kid who has been taught to use AI will paste that code and ship it. A kid who has been taught to build with AI will run it, read it, test the edge, check the reference, and only then put her name on it. We rehearse this. We give kids broken AI output on purpose and watch them find the bug. The mentor sits next to them and says what made you suspicious of that line? The habit gets built early because we built it on purpose.

This is the single most durable skill in the room. The model will be a different model in six months. The verify habit will still be the verify habit.

What this looks like on a Tuesday

A nine-year-old wants to build a card game his sister will play. He sketches the rules with a mentor. He uses an AI tool to scaffold the first version of the code. He runs it. It works except his sister can play an illegal move on turn three. He reads the code, finds the rule the model forgot, fixes it, ships v2. His sister plays it. He puts the screenshot on the shelf.

That is the entire pedagogy in one Tuesday. Foundation, multiplier, verify, ship.

It scales. The fourteen-year-old is doing the same loop with a tuned agent and three paying users. The eight-year-old is doing it with a Pokédex and a sibling. Same loop. Same room.

That is what we mean.