Idea: Recursively generate code until unit tests pass with as much margin as possible.

Procedure

• Create a design code, a basic code repository and a set of well defined unit tests
• Hand this to an AI agent with the ability to write code, run unit tests and make commits to the repo
• Instruct the agent to implement the design doc and run the unit tests
• Instruct the agent to recursively update the code, run the unit tests until they pass with the highest margin possible (probably good to have it stop after something like 10 or 100 attempts because cost)

Applications

• Code that handles hardware or radios where physical margin matters
• Code that requires scale like web services
• Code that requires performance like video encoders