How to measure AI code quality?

I’m also less sold on plans as time goes on and more sold on defining what it needs to do and being able to verify that objectively. The main “plan” I use now is to just specify the top level abstractions and that’s it. The more experience I have doing full harness builds the more I think you should define what it needs to seo at the boundary and let the architecture emerge from the build and formalize later if necessary.

1 Like