I've been working on scaling my throughput for AI development and hitting a wall with git worktrees. The problem is worktrees are great for isolating edits and reads for code but bad for running code. The more agentic coding I do the more obvious it is to me that you get the maximum value by collecting runtime evidence.
I built Moo, https://github.com/heyito/moo to help with this but I'm curious about how other people are preventing their agents from stepping all over each other.
Top comments (0)