With crews, one lead agent delegates to others, so accountability gets blurry fast. How do you track which human is responsible when agent A hands work to agent B?
One approach I’ve seen (in Sharkly.ai, disclosure: I work on it) records a Triggered by field on every run that traces back up the chain to the person who started it.
For those running CrewAI in production: do you log delegation chains? Do you enforce human approval before a crew’s output is used?