Simon Willison says AI has crossed a new threshold in making playable games

AI models are beginning to make games that people enjoy playing, not merely convincing-looking prototypes, argues developer Simon Willison. His examples suggest that game design itself, long a stubborn human part of the process, is becoming easier to hand to a model.

Watch Desk analysis

What happened

In an essay published on 28 October, Willison describes four browser-based games he built with AI assistance: a spot-the-difference game, a fantasy RPG, a real-time strategy game and a pumpkin-farming simulator. He says his work on each involved no more than four hours of his own effort, alongside longer stretches of model work.

The strongest example is Battle Drones, a multiplayer strategy game he says was transformed after he asked Claude Opus 5.5 to make it “a LOT more fun”. It now has multiplayer, seven maps, replays and a random map generator. Willison also says the model’s game-design ability, rather than simply its ability to produce code or graphics, is what changed his view.

Willison presents this as a personal account, not a controlled comparison. He says a version of the fantasy RPG built with ChatGPT GPT-6 Pro looked good but did not play as well, and that his pumpkin-farming game still needs work.

Why it matters

Games combine code, art, music, dialogue and design. If one person can use AI tools to make a playable, distinctive game with little technical experience, the barrier to experimentation falls sharply. That could bring more people into game-making, while raising familiar questions about the value of creative work and the flood of AI-made material.

The useful distinction is between making something playable and making something people want to play. Willison’s examples offer a concrete account of models helping with both, but his own playtesting and repeated prompting were part of the process. “Make it more fun” is a striking prompt; it is not yet a substitute for taste or iteration.

Our read

This is a substantial practitioner’s demonstration, not proof that AI can reliably design great games on command. The most interesting claim is that models are starting to help with the hard-to-formalise bit: the play loop. That is worth watching, even if the industry’s next challenge may be sorting delightful experiments from a very large pile of merely playable ones.

What to watch

• Whether other developers report similar results across different models and genres.
• How much playtesting and human direction strong AI-assisted games still need.
• Whether new ways of discovering and evaluating AI-made games emerge.

Discussion spark: If AI makes it easy to build a playable game, should we judge the result mainly by how much fun it is, or by how much creative work the person behind it contributed?

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.