anyone actually running a multi-agent "team" that keeps going without you?

most setups i see still die after one chat — you spawn more agents, then babysit every handoff. curious what’s working in practice for long-running agent graphs (lead + workers + approvals), not just "better prompt". saw lauren tan’s spacexai workshop / grok bot always-on team writeups floating around — curious if people are doing similar with claude code / desktop agents without it turning into babysitting. not selling anything, just want war stories / failure modes.