The workflow involves multiple AI agents, primarily through claude code and o3, operating on git-worktrees to generate and verify code. The process begins with the creation of a high-level plan by o3, followed by execution through sonnet models. Plans are constructed to ensure tasks are clearly defined, allowing the agents to continuously validate and improve their output without manual intervention. The focus remains on fixing input parameters to influence outputs correctly, making adjustments based on learned issues to enhance future executions and maintain a self-improving system.
The factory improves itself by generating code, executing plans, and verifying results, with any issues being fed back to enhance future plans.
Adjustments are made to the plan, prompts, or agent mix to ensure future outputs are correct by construction, rather than manually fixing errors.
Using claude code and o3 enables a systematic workflow where complex tasks are broken down into plans that are executed and self-verified.
Inspired by Factorio, the process mirrors building a self-sustaining factory of AI agents that continuously produce and enhance their own code.
Collection
[
|
...
]