Ok, tried `claude code` sub agents on a fullstack system. Made: 8 `specialists` 1 `troubleshooter` 1 `architect` 1 `project manager` I made the PM orchestrate the other sub-agents, then migrated the agents from claudes' internal TODO list to a project based system for state permanence between machine resets First test, it 1 shot an issue with fresh context, 1 line of prompt with a fresh set of system logs. The same issue was spiralling yesterday over many attempts. So glad it's officially supported now. Next up is (thanks @0xz80) is hooking in mcp and Gemini to be the PM and Architect rather than claude.