What we did
My kids each invented a game. They decided what should happen, what the player could do and what made the game fun.
They did not start with a prompt or a template. They started by talking through the ideas in their own words.
What happened
The first builds were playable, but they were not finished. Playing made the children notice unclear moments, missing feedback and rules that did not match the idea in their heads.
They could judge the result and direct another version without needing to understand the code.
How we did it
The tools were practical and visible. Nothing about the workflow needed to be hidden from them.
- 01Kids come up with the game idea
- 02They explain it verbally
- 03WhisperFlow transcribes it
- 04ChatGPT turns the idea into a project spec
- 05A coding agent builds the game
- 06Kids playtest
- 07Kids give feedback
- 08The coding agent changes the game
- 09Repeat
- 10Deploy
What I learned
The interesting part wasn't that AI could code the games. It was that implementation stopped being the bottleneck.
The children were still responsible for imagining, describing, judging, changing and iterating. AI made their decisions visible and playable, but it did not make those decisions for them.
AI did the implementation. The ideas, decisions and judgment stayed with them.
What I'd do differently
I would make the first build even smaller. A rough version gives children something concrete to react to, and their judgment becomes clearer once they can play.
I would also capture more of their exact language. The corrections often reveal more than the finished game does.