I let my 6-year-old design her dream video game with AI. Here's what she made.

My six-year-old asked if she could make a video game on a Saturday morning. She had been watching a friend play Minecraft and decided this was now her thing. She is not yet a confident reader. She cannot spell dragon. She can absolutely tell you, with extreme conviction, exactly what kind of video game she wants.

I tried ScratchJr first. We sat down with the iPad. I read the labels out loud. We dragged blocks around for about fifteen minutes. She built a kitchen scene with a cat character. We made the cat move when tapped. She played the result for maybe two minutes — and then asked if the mice could be different colours.

I started fiddling with the sprite editor to show her how. By the time I had it open, she was draped over the back of the sofa, picking at a hangnail. I cannot blame her. The reward was real but it was buried under fifteen minutes of process she did not choose, did not understand, and could not drive herself.

The inversion

That afternoon I prototyped what eventually became BrainLab Games. The premise: same kid, same idea, but the input is a sentence and the output is a real playable browser game in about sixty seconds. The kid is the director. The AI is the studio.

The first prompt my daughter typed (with me reading and pecking on her behalf) was eight words long:

a cat chasing mice through a kitchen

About seventy seconds later, a cat was chasing mice through a kitchen. She played it for ten minutes. Then she said, can the cat be purple. I typed make the cat purple. Thirty seconds later, the cat was purple. She played another five. Then she asked for orange mice. Then dinosaurs instead of mice. Then mice and dinosaurs. Then she wanted the kitchen to be a spaceship. The whole afternoon went like that.

What changed

The thing that worked was not the AI getting fancy. It was the loop being short enough that her attention could stay in it. ScratchJr's loop is imagine → drag → drag → drag → run → adjust → run → finally play. The blocks are well-designed. The kid is held back not by the tool but by their own keyboard skills, reading speed, and patience for sequencing. By the time the reward lands, half the kids are gone.

The new loop was imagine → say → play → "I want this different" → say → play. The act of articulation became the creative act. She was learning something — but it wasn't programming. It was specification. She was discovering, with every iteration, that the words you choose change the thing you get. Purple produces purple. Make it harder produces harder. Make it weirder produces, hilariously, much weirder. This is a skill people pay product managers six figures to have.

Six months in

She is now seven. She has made forty-three games in our family account. Her younger sister, who is four, has made fewer (she usually narrates and lets a parent or older sibling type) but she plays the games her sister directs — including a long-running franchise about a small dog who has to escape from broccoli — for hours.

Things I did not expect:

• The games she makes are, on average, better than the ones she chooses to play on YouTube. She is more invested because she made them. She is more willing to lose, because she can change the rules.
• Her vocabulary has noticeably widened in the categories the AI keeps surfacing — parallax, combo, boss, health bar. Words she would not have met for years otherwise.
• She has invented a thing called a remix, where she takes a finished game and asks for one specific change. ("Same game, but the cat is a teacher now.") This is, of course, exactly how all professional game design works.
• She wants to share the games with her cousins. The cousins play them. The cousins want their own. The hottest thing on family WhatsApp right now is a fourteen-year-old's bullet-hell game starring an angry cucumber.

Things I worried about, in retrospect

Was this just letting a screen do the work for her? The honest answer is no, and the test is what she does when the game lands. She does not passively consume. She rejects, edits, iterates. The screen is the canvas, not the babysitter.

Is she "learning to code"? No. ScratchJr is for that. If you want your kid to learn the shape of programming, install ScratchJr and sit with them. They are different goals. I think both matter; I think mine, in this household, currently cares more about creative iteration than about syntax.

What about screen time? She does the same thirty-to-sixty minutes a day she used to do on YouTube. She just does this instead. The AAP's most recent guidance distinguishes between creative and passive screen time and says creative is fine in much larger doses. This sits firmly in the first bucket.

Is the AI ever weird? Yes, occasionally. Once it gave her a game where the cat was on fire for no clear reason. She thought that was the funniest thing that had ever happened to her. We sometimes screenshot the misses; they are usually the most-shared output from any given week.

The thing this product really is

I started this thinking I was building a kids' creative tool. I think I was actually building something narrower and more useful: a tool that closes the gap between a child's idea and a playable result, fast enough that the idea is still warm. That gap is, in my experience, where most of childhood creative ambition goes to die. We hand kids paint and they get a flat painting. We hand them coding tools and they get a tower of blocks they can't yet run. We hand them this and they get the thing they were thinking of, on the screen, within their attention window.

That is the whole thesis. The rest is implementation.

— Louis, parent of two, builder of one game-maker.