Creativity · Agent Protocol
CrewAI Hierarchical Process
CrewAI lets you organise multi-agent 'crews' under two processes: sequential, where tasks run one after another, and hierarchical, where a manager agent plans work, assigns it to crew members, inspects their outputs, and re-assigns if needed. Hierarchical is the right choice when the task doesn't decompose cleanly up-front and a manager's judgement is part of the loop.
Protocol facts
- Sponsor
- CrewAI
- Status
- stable
- Spec
- https://docs.crewai.com/concepts/processes
- Interop with
- LangChain, A2A, OpenAI, Anthropic, MCP
Frequently asked questions
When should I use hierarchical vs sequential in CrewAI?
Sequential is fine when you can list the steps up-front — do A, then B, then C. Hierarchical wins when the manager needs to adapt — assign one task, look at the output, decide whether to rerun, redirect, or move on.
Does CrewAI's manager need to be a specific model?
You can pick the model, but best practice is to use a stronger model for the manager (it does the planning) and lighter models for worker agents doing focused sub-tasks.
Is this hierarchical in the multi-level sense?
CrewAI's 'hierarchical' process is a two-level shape — one manager over N workers. Deeper trees are possible by nesting crews, but the built-in process is a single level of delegation.
Sources
- CrewAI — Processes documentation — accessed 2026-04-20