A "make-me-a-game" button would be the most-clicked feature in any kid-facing AI studio. It would also collapse the construction loop the studio runs on. Self-determination theory predicts the kid’s autonomy and competence collapse with it. Csikszentmihalyi’s flow research predicts the kid never finds the optimal-challenge zone. Forty years of motivation research point at the same shape: scoped decisions, not big ones, are what keep kids engaged with the work.
The temptation is real.
We should start by admitting how appealing the idea is. A kid sits down. There’s a big friendly button on the home screen. They type "a game about a dragon who collects stars" and twenty seconds later there’s a game about a dragon who collects stars. They show their parent. The parent is impressed. The kid is delighted. Engagement metrics are great. The screenshot ships well on social.
If we shipped this feature it would probably double our trial-to-paid conversion. We have considered it. We’ve sketched it. We have working prototypes that do exactly this thing. We didn’t ship them, and we won’t.
The reason is that we’ve already written about the principle in why we don’t outsource creativity to AI, and this post is the practical sequel. There we explained the framework. Here we explain what actually happens in the kid’s session, minute by minute, when you press a button like that, and why the alternative (a scoped wizard) is so much better even though it asks more of the kid.
What happens when the kid presses it.
Watch a kid use a "make-me-a-game" feature for a full session. The pattern is consistent across tools that ship this shape of interaction.
Minute one: the kid types a prompt. Maybe "a dragon who collects stars". Maybe "Fortnite but with cats". The system thinks for a few seconds and produces a thing.
Minute two: the kid plays it. If it’s good they say cool. If it’s not quite right they shrug and re-prompt. The cycle takes about ninety seconds.
Minute ten: the kid has produced five or six different games. None of them is theirs. They didn’t make any of them. They typed a sentence and the machine made the thing. They can’t open the game up and change one detail; the game came as a unit and they have no purchase on its insides.
Minute twenty: the kid has run out of prompts. The novelty is wearing off. They start typing increasingly outlandish requests trying to get the system to surprise them. The system either does (briefly) or doesn’t (and they bounce).
Minute forty: the kid closes the tab. The session is over and they didn’t take anything with them. There’s no project they want to come back to tomorrow because there’s no project they were building. They were consuming.
This isn’t a hypothetical. It’s what shows up when you instrument the analytics on tools that ship the make-me-a-game shape. The session-one engagement looks great. The session-three engagement is gone. The kid never builds a maker identity because there’s no artifact carrying one back to them.1
What happens with a scoped wizard.
Now watch a kid use Tell and Show, where the only AI surface is a scoped wizard like "add a character" or "make this scene scarier".
Minute one: the kid picks a track. Game. Story. Site. Movie. They’re looking at a real, running, almost-empty project. There’s a character on the screen and it’s walking around. It’s the kid’s artifact already because they picked the track and they’re looking at it.
Minute three: the kid runs their first wizard. "Add a character." A small form appears asking the kid for the character’s name, what they look like, and what they should do. The kid has to make those decisions. The AI proposes a scoped change. The ChangeDisclosure card shows what changed in plain English. The kid presses Keep.
Minute six: the character is in the game and the kid is poking at the controls. They run another wizard. "Add an enemy." Same shape; same scoped autonomy; same small decision the kid has to make.
Minute thirty: the kid has run six wizards and made twelve micro-decisions. The project has their fingerprints on it. They can point at specific things and say "I picked that color" or "I chose that name". They know how to undo something. They know the AI will propose and they will decide.
Minute sixty: the kid’s parent walks in. The kid wants to show them the game. The kid is using first-person ownership language ("I made this part") because the studio architecture made that the only honest language for what happened. They want to come back tomorrow because the project isn’t finished and they’ve been steering it.
The difference between the two sessions is not the AI’s capability. The same underlying model can power either shape. The difference is what the kid was asked to do. In the first session the kid was asked to specify. In the second the kid was asked to decide. Specifying is a one-time act. Deciding is a habit.
Why scoped autonomy wins.
The motivation research on this is pretty unanimous and it goes back decades. Edward Deci and Richard Ryan’s self-determination theory predicts that humans need three things to be intrinsically motivated: autonomy, competence, and relatedness.2,3
A make-me-a-game button is a curious case because at first glance it seems to give the kid more autonomy, not less. The kid is in charge. They picked what to make. But SDT is specific about what autonomy means, and it doesn’t mean "the AI did what you said". It means "you experience yourself as the cause of the action".2 When the AI hands the kid a finished thing, the kid is the cause of the request, not of the action. Autonomy is hollow.
Specifying is a one-time act. Deciding is a habit. The studio is built to grow the habit. On the difference between a button and a wizard
Competence is the second leg, and the make-me-a-game button collapses it almost completely. The kid isn’t getting better at anything during the session. The AI is getting better at predicting what the kid wants, but that’s the AI’s competence, not the kid’s. Scoped wizards do the opposite: every keep / review / undo decision the kid makes builds their taste, and by the tenth one they can feel the improvement.
The third research thread is Mihaly Csikszentmihalyi’s flow research, which describes the conditions under which a person becomes deeply engaged with a task.4 Flow happens when the challenge of the task matches the skill of the person. Too easy and you’re bored. Too hard and you’re anxious. The kid pressing a make-me-a-game button has no task; the AI did it. There is no challenge curve. Flow can’t happen. Scoped wizards do the opposite: each decision is small enough to be doable and meaningful enough to matter, which is precisely the conditions Csikszentmihalyi described.
The fourth thread, and the one that closes the loop, is Mitchel Resnick’s four Ps from the MIT Lifelong Kindergarten group: Projects, Passion, Peers, Play.5 The framework asks: is the kid working on a Project they have Passion about, with Peers, in a spirit of Play? A make-me-a-game button satisfies none of the four. The kid didn’t do a project; they ran a query. They have no passion because they didn’t struggle for it. They have no peers because there’s no work product to show. Play is replaced by demand. Scoped wizards keep all four legs intact.
The design constraint.
So that’s why we don’t ship the button. It would be the most-clicked feature. It would also be the most-regretted, because the kid would have a great twenty minutes and lose interest by the third session, and we’d have signed up parents for software their kid stopped opening.
The constraint we keep returning to is: every AI surface in the studio operates on a project the kid is already making, and every change is scoped, named, and reversible. The kid is the author. The AI proposes. The kid decides. The pedagogy only works inside that shape.
It’s a slower onramp than a make-me-a-game button. The first session is less impressive. The conversion rate from the home page is lower. We can live with that, because the second session is better than the second session anywhere else, and the third session is when the kid starts coming back on their own.
If you want the deeper version of the argument, the parent post is why we don’t outsource creativity to AI. If you want to see the constructionist tradition this design sits inside, start with why making is learning. And if you want to skip the reading and just see what a kid actually does in the studio, the demo at tellandshow.ai/try takes about three minutes.
References
- Yasmin Kafai & Quinn Burke, Connected Code: Why Children Need to Learn Programming, MIT Press, 2014. On the difference between consuming digital media and producing it, and the long-run engagement gap between the two.
- Edward L. Deci & Richard M. Ryan, Intrinsic Motivation and Self-Determination in Human Behavior, Plenum, 1985. The foundational text on self-determination theory.
- Richard M. Ryan & Edward L. Deci, "Self-Determination Theory and the Facilitation of Intrinsic Motivation, Social Development, and Well-Being," American Psychologist, 2000. The most-cited synthesis of SDT.
- Mihaly Csikszentmihalyi, Flow: The Psychology of Optimal Experience, Harper & Row, 1990. The original articulation of the flow concept and the challenge-skill balance.
- Mitchel Resnick, Lifelong Kindergarten: Cultivating Creativity through Projects, Passion, Peers, and Play, MIT Press, 2017.