I set up an orchestration in Claude Code with 6 agents - planner, coder, reviewer, tester, documenter, and committer. Each one does exactly what the name says.
To make it work I had to define each role clearly - what they own, where one stops and the next begins. When the role definitions were vague, the whole thing broke down. When they were clear, I could ship a new feature in 3 prompts.
What I noticed -> that's exactly what happens with real teams. Unclear roles create friction. Clear roles create flow.
I've been managing teams for years and I think I just learned something from AI about how to do it better.
Has anyone else noticed something like this - where working with AI taught you something about working with people?