Every game we publish came from a single written brief and one build pass. Here is the honest scorecard.
The experiment
We wrote a short prompt for each game, handed it to an AI coding agent, and kept the first build. Then we ran scripted checks and played them ourselves. That produced 12 original games and 20 classic board and puzzle games, from a daily word game to a full chess engine. The prompts are public on each game page.
What one prompt got right
- Rules. Checkers, Reversi and four in a row ran correctly first time: compulsory captures, passes, winning lines.
- Look and feel. A neon arcade style held across every game once the prompt named it.
- Housekeeping. Pause, game over, best score saved: all there when the prompt asked.
What it needed help with
- Verification. Chess is the example: the moves look right until a single edge case breaks castling or en passant. We tested the move generator against the standard move-count benchmarks before publishing it.
- Touch and phones. Our scripted tests used simulated screens. A real phone test is a different thing, and that's why a person played them before we called them done.
- Feel. Speeds, difficulty and balance are judgement calls. A first build is a first draft.
What one prompt can't do
It won't make a game you can't describe, and it won't know your taste. A prompt that says "make it fun" returns a competent average. The prompts that worked named the controls, the scoring and the way a run ends. A single prompt got us working small games. It didn't get us great ones, and we don't claim otherwise.
The takeaway
For a small, well-defined game, one good prompt gets surprisingly far. The work moves from writing code to specifying clearly and checking carefully. See real prompts or make your own.
What we would do differently
If we ran the experiment again, we'd write the controls and the loss condition into every prompt on day one, and we'd test on a physical phone before calling anything finished. Those two habits account for most of the problems we found. We'd also keep a note of every change we made after the first build, so the iteration count on each game page is as honest as it can be.
Questions people ask
Did every game work first time?
They ran, and the scripted checks passed. A human playtest then looked for feel, balance and phone issues.
Where are the prompts?
On each game's page under the player.
Does a longer prompt make a better game?
Not by itself. Clear controls, scoring and a loss condition matter more than length.
Keep reading
- How to Make a Game With AI: A Beginner Guide
You can have a playable game in an afternoon. Here is the route that works, the mistakes that waste a weekend and how to put the result online for free.
- Vibe Coding Prompts for Games: Real Examples
Most advice about prompts is theory. These are the actual prompts behind games you can play on this site, with notes on why each one worked.
- What are vibe coded games?
A practical guide to building and testing games with AI.