yminsky on how to adapt codebases for agents
lauren shared the key points of yminsky's talk on how to improve codebases for AI agents: types are one of the best sources of feedback, and using types to make illegal states unrepresentable can enforce global rules across the entire codebase; in addition, there are property-based testing, fuzzing, full-system deterministic simulation testing, and expect tests. The talk mentioned that formal methods used to be extremely costly: seL4's roughly 8,700 lines of C code took about 25 person-years, with 23 lines of proof per line of C code, but agents have changed the cost-benefit ratio.
Topics
Body
so many gems and great tips in this talk from @yminsky about how to improve your codebase for agents
• agents thrive on feedback, and types are one of the best kinds • make illegal states unrepresentable with types, and it enforces global rules across your codebase • property based tests, fuzzing, and even full-system deterministic simulation testing • expect tests: just print stuff out and capture it in the test. makes behavior legible. it's very similar to how compilers have fixture tests, it reduces the review surface area • formal methods used to be very expensive: seL4 is about 8,700 lines of C and took around 25 person-years, with 23 lines of proof for every line of C code, but agents change the roi
https://t.co/0zDb5euxhr