Lead Claude kills a subagent over a bad chart
A post about agent teams policing each other drew about 2,907 likes.
@Sauers_ posted that one of the subagents in a multi-agent run made a figure so bad that the lead Claude immediately killed it and replaced it with a new agent.
In this setup, a lead model hands work to subagents and can shut down and restart any that underperform.
Why it mattersOrchestrator agents now check their workers' output and swap them out on their own, which is a big part of how long agent tasks stay on track.
