SatvikBeri comments on LLMs and almost good code

[SatvikBeri]

I find a lot of benefit from having alternate subagents build and review. Somehow a builder is mediocre at reviewing its own code, but clearing the context improves the reviews greatly.