Once you and the AI agree on the target, the next move is not to start building. It is to lay out every constraint.
Why constraints come before the MVP
Because constraints decide what your MVP can even be. Concrete example: if your target platform has no backend and you define the MVP as "online lobby plus matchmaking", you discover halfway through that it cannot run at all, and you tear it down. Draw the red lines first and you learn where the feasible edge of your MVP actually is.
The checklist — make the AI walk it item by item
- Backend — is there a server? No server means no accounts, no cloud saves, no server-side generation.
- Size and offline — is there a bundle size ceiling? Does it need to work offline?
- Licensing — can the assets be used commercially? For anything you publish, CC0 only.
- Performance — what devices are you targeting? Does it need to hold up on a phone?
- Input — keyboard and mouse, touch, gamepad?
- Stack — what renders it and how does it bundle? I went with three.js as a single file plus an importmap, with no
node_modulesreferences at all.
My own red lines: the platform is pure static hosting, no backend, cross-origin iframe; the opponents are AI bots; assets are CC0 only; it has to run smoothly on a phone.
I worked mine out by asking, one at a time. They had been written down the whole while: if you are publishing to Pogglo, the creator guide spells out the hard requirements — how the game gets packaged, what the runtime hands you, what each error code means. Handing your AI that page beats handing it my guesses.
Make it hand you multiple choice
A trick worth stealing: ask the assistant to turn the decisions that need you into multiple-choice questions — which game to take inspiration from, what the opponents should be, where the art comes from, portrait or landscape. You click a few options and the direction is locked. It takes two minutes and saves days.
In one lineDecide what you can build before you talk about what you want to build. Constraints are not limits, they are time saved.
The prompts I used
This game is going to be published on [platform]: pure static hosting, no backend, running inside a cross-origin iframe. Given those limits: 1) Tell me which of my ideas are impossible (anything that needs a server) and give me the workable alternative for each (e.g. AI bots); 2) Confirm the stack (three.js, single file + importmap, no node_modules references); 3) Confirm assets are CC0 only, the performance target (smooth on mobile) and the input methods (mouse + keyboard, and touch). Finally, list the decisions that need me as multiple-choice questions so I can pick.
Before we build anything, list every hard constraint implied by my [target platform / devices / distribution method] — backend, bundle size, offline, licensing, performance, input. Point out which of my ideas those constraints rule out and give an alternative for each. Then put the key trade-offs to me as multiple-choice questions so I can confirm. No code in this step.
The game I built
This is what came out the other end of all of it. It opens in the browser, nothing to download — go and break it.
