I am currently trying to build a complex research workflow and heard that is the go-to framework for orchestrating multiple agents. However, I’m struggling with how to properly define backstories so agents don't overlap. Has anyone successfully managed to get a 'Manager' agent to delegate tasks effectively without getting stuck in a loop?
3 answers
To get delegation right in , you need to ensure the allow_delegation parameter is set to True specifically for your manager agent. The secret sauce is in the backstory; you must explicitly define the manager's authority over the 'Researcher' and 'Writer' agents. In my experience, loops usually happen when the tasks are too broad. Try breaking them into granular chunks. I’ve been using this setup for my content marketing agency since late 2025, and it significantly reduced my manual oversight once the roles were strictly defined in the process.
Are you using the sequential or hierarchical process flow for this specific project? I’ve noticed that hierarchical flows require a much more robust LLM like GPT-4o to handle the delegation logic without tripping up
I found that adding a 'Reviewer' agent helps a lot. It acts as a final gatekeeper before the task is marked as complete, which catches those delegation errors early on.
Adding a reviewer is a solid tip! I did the same with my crew and the output quality for my SEO blogs improved by at least 40 percent.
I’m actually using the sequential flow right now, Karen. I thought it would be simpler, but the agents seem to ignore the previous output sometimes. Should I switch?