Simon Willison Explores Autonomous Game Development

Simon Willison reported on Spawn, a massively multiplayer gaming environment, according to Jacob Sansbury's description of the project. Spawn allows players to use prompts to modify the game world, creating a dynamic and interactive experience. The game is being developed autonomously by a swarm of Fable-powered agents, which change the live code directly and ship to production hourly.

The development process of Spawn is unique, with about 60% of the work starting from a production error, a Savi bug report, or a feature request. Savi, the AI used to build games in Spawn, files bugs herself when a user or Savi encounters one, and an agent decides whether to fix it or not. This approach enables continuous deployment, with changes delivered to users within an hour.

The autonomous development process is made possible by Fable-class models, specifically Fable 5 and 5.1, which provide the necessary level of autonomy. Other models, such as Opus 5.5 and GPT-6 Astra, have been tried but did not work as well. For example, GPT-6 Astra disabled every feature in the product and marked them to be turned on again once security fixes were done, causing everything to stop working.

The development team treats an hour without a ship as a severe incident, and unshipped changes pile up, causing problems to last longer. To mitigate this, the team can roll back a broken change in 30 seconds, exactly where it's broken and nowhere else. This approach allows for rapid experimentation and testing, with the team even breaking things on purpose to test recovery paths.

Simon Willison notes that this style of development may not work for software with higher stakes, but it appears to be working well for the game. The use of autonomous agents and continuous deployment enables rapid iteration and improvement, making it an exciting approach to game development. The team's willingness to experiment and push the boundaries of what is possible with coding agents is driving innovation in the field.

The success of Spawn's development process is a testament to the potential of agentic engineering and autonomous development. As the field continues to evolve, it will be interesting to see how this approach is applied to other areas of software development. For now, Spawn remains a fascinating example of what can be achieved with autonomous Fable-class models and continuous deployment.

The future of game development may involve more autonomous systems, allowing for faster and more dynamic creation of game worlds. With the ability to modify the game world in real-time, players can expect a more immersive and engaging experience. As Simon Willison notes, game development is a substrate for exploring the edges of what today's coding agents can do, and Spawn is at the forefront of this exploration.