My sessions are significantly better if I have one agent investigate to write up a PRD/Plan and cycle that with feedback two or three times with me (usually Gemini or Claude plan mode).
My sessions are significantly better if I have one agent investigate to write up a PRD/Plan and cycle that with feedback two or three times with me (usually Gemini or Claude plan mode).
Another tip a friend offered is: Claude is intentionally optimized to minimize token usage. You must consistently instruct it to do anything eager (read other files) or it will miss cases.
Another tip a friend offered is: Claude is intentionally optimized to minimize token usage. You must consistently instruct it to do anything eager (read other files) or it will miss cases.