Why making is learning — and why AI didn’t change that.

The pedagogy isn’t new.

In 1980, Seymour Papert published Mindstorms, a book about kids and computers and the kind of learning that happens when you give a child a system they can build inside.¹ Papert had been at MIT for almost two decades by then, working alongside Marvin Minsky and others, but the book wasn’t about artificial intelligence. It was about a small, slow-moving robot called the LOGO turtle that a kid could tell to walk in a square or draw a flower. The argument behind the book had two parts.

First: kids learn best when they’re building something. Not when they’re being told facts. Not when they’re doing exercises that someone else designed. When they’re trying to make a thing exist in the world that didn’t exist before.

Second: the thing they build matters. Papert called this constructionism, a deliberate one-letter variation on Jean Piaget’s constructivism.² Piaget had argued that knowledge is constructed inside the learner’s head. Papert argued the construction is sharper, and lasts longer, when the learner is also building something outside their head. A physical or digital artifact they care about.

The two arguments compound. Kids build something that matters to them; in building it, they construct the understanding that lets the thing exist; the understanding is durable because it’s entangled with the artifact. The artifact is the proof of the learning. You can pick it up. You can run it. You can show your friend.

Why most edtech ignores this.

Constructionism is harder to operationalize than instructionism. A teacher telling a learner facts scales to thirty kids in a classroom. Constructionism doesn’t. A kid building a thing they care about needs someone who can hear what they’re trying to do, point at the next move, and step back when the kid wants to keep going. The role is closer to a craft mentor than a teacher.

Vygotsky had a name for the role. He called the space between what a learner can do alone and what they can do with help the zone of proximal development.³ The mentor’s job is to scaffold the learner across that gap. Not by doing the work. By holding it at just the right reach. Too easy and the kid isn’t learning. Too hard and they disengage.

The mentor’s job is to scaffold the learner across the gap. Not by doing the work. By holding it at just the right reach. Vygotsky’s zone of proximal development, restated

This is the move most software for kids has been trying and failing to do for forty years. Scratch comes closer than most.⁴ Code.org tries. MakeCode tries. The good ones reduce the friction between "I have an idea" and "the idea is running." The great ones, LEGO and Minecraft and the original Hypercard, do this and also let the kid’s ambition exceed their technical reach. They expand the zone.

What they couldn’t do, until recently, is hold a real conversation with the kid about what to build. Scratch knows what blocks are available. Scratch can’t hear "I want this character to feel scary" and propose three ways to make that happen. The kid’s ambition had to be translated into the tool’s vocabulary, and the translation was the kid’s problem.

What AI changes, and what it doesn’t.

People miss this when they call AI tools for kids "the next Scratch." They’re not. They’re the next Vygotskian scaffold: a partner in the zone of proximal development that can listen, translate, propose, and step back. Scratch was the medium. AI is the partner.

That distinction matters because it predicts which AI tools actually teach. The ones that act like instructionism with extra steps, where the AI hands the kid a finished thing on request, collapse the construction loop. The kid never built anything. They received it. The artifact isn’t theirs. The understanding isn’t durable. Papert’s argument fails.

The ones that act like constructionism with a better partner, where the AI proposes a change, the kid sees what would happen, the kid keeps or reviews or undoes, sharpen the loop. The kid is still building. The AI is just close enough to the kid’s ambition to keep them moving. Papert’s argument holds.

This is the dividing line Tell and Show is built around. The studio is a constructionist environment that happens to have an AI inside it. The wizards don’t make games for kids; they propose changes to games the kid is already making. The chat doesn’t deliver answers; it surfaces moves the kid then decides about. The ChangeDisclosure card is visible because Papert’s constructionism only works when the kid can see, and own, each change.

What we got from the existing research.

We didn’t invent any of this. The studio borrows ideas that have been tested, and in some cases re-tested, in ed-tech research over multiple decades. The shortest version:

• Papert — constructionism. Kids learn by building artifacts they care about. The artifact is the proof.^1,2
• Piaget — constructivism. Knowledge isn’t transmitted; it’s built. Constructionism stands on this.⁵
• Vygotsky — zone of proximal development. Kids reach further with the right scaffolded partner.³
• Resnick — the four Ps. Projects, Passion, Peers, Play. The MIT Lifelong Kindergarten group’s framework for what makes kid-creative environments work.⁴

Each of these names a decision we made about the studio. Papert tells us the kid needs to build, not be told. Piaget tells us the knowledge has to be assembled by the kid, not delivered. Vygotsky tells us the AI’s job is to scaffold, not to solve. Resnick tells us the kid needs a project they care about, the passion to keep going, peers who see it, and a playful spirit while making it.

What our generation has to figure out.

The honest answer is that we don’t yet know all of what AI as a constructionist partner does to a child’s long-term learning. The framework predicts the loop will work; the research base is still being built. We’re three years into a new medium, not thirty.

What we do know, from a growing line of AI-literacy research at MIT, at Stanford’s Graduate School of Education, and at places like the AI for Education group at Wharton, is that visible AI does better at teaching than opaque AI.⁶ Kids who watch the model work, who name what it’s doing, who debate with it, end up with better mental models of how AI actually behaves. Hidden AI, the kind that hands you the answer, produces kids who treat AI as magic.

The studio is built around that finding. The tool-trace pane, the ChangeDisclosure card, the Keep / Review / Undo decision row. They exist because the research and our own cohort observations agree: kids learn AI by watching it work, and by deciding what to keep.

Papert ended Mindstorms by writing that the computer was, for the kids he’d watched, "a personal medium." Forty-five years later that’s still the right phrase. Tell and Show is one attempt to make AI a personal medium for the next generation of kids. The research that came before us tells us how to do it carefully. We’re trying to honor it.