build1 publisher
A three-agent CrewAI run spent 44.6 of its 118 seconds inside coworker tool calls
CrewAI hands a delegating agent two functions. Calling one starts a fresh model call as the named coworker. In this run the expensive reviewer caught a price error it had no source to check.
Publishers:dev.to
Reality
- Evidence55
- Adoption12
- Hype gap−10
- Incentives30
- Confidence58