CrewAI Hierarchical Content Crew

What This Builds

This recipe uses CrewAI’s hierarchical process, which emulates a corporate hierarchy: a manager agent oversees planning, delegation, and validation, while specialist agents do the work. Crucially, tasks are not pre-assigned — the manager allocates work to agents based on their capabilities, reviews outputs, and decides when a task is complete.

The concrete use case is a content production crew. A manager coordinates:

• a researcher that gathers and synthesizes source material,
• a writer that drafts the piece, and
• an editor that reviews and requests revisions.

The manager loops: delegate, inspect, send back for fixes, and only finish when the output meets the bar.

The Stack

• CrewAI — define Agents with roles/goals and allow_delegation, define the deliverable as a Task, then assemble a Crew(process=Process.hierarchical, ...). Hierarchical mode requires either a manager_llm (CrewAI auto-creates the manager) or a custom manager_agent.
• LLM providers — a fast model on GroqCloud for the high-volume worker turns and a stronger model (via OpenRouter, or directly) for the manager’s planning and validation.
• Qdrant (Cloud free tier) — optional vector store behind a research tool so the researcher can retrieve from a private knowledge base.

Step-by-Step Outline

1. Define specialist agents. Researcher, writer, and editor, each with a focused role, goal, backstory, and allow_delegation set appropriately. Give the researcher a retrieval tool backed by Qdrant if you want grounding.
2. Define the deliverable task(s). Describe the desired output and expected format; do not hard-assign which agent does what.
3. Configure the manager. Set process=Process.hierarchical on the Crew and provide manager_llm="..." (or a custom manager_agent with task-management skills).
4. Kick off. Run the crew; the manager plans the work, delegates to specialists, reviews their outputs, and iterates until satisfied.
5. Tune. Watch for the common pitfall where defining only manager_llm (and letting CrewAI build the manager) behaves more reliably than an under-specified custom manager; add tools to the manager only when you confirm your CrewAI version supports manager tools.

Why This Shape Works

The hierarchical process centralizes planning and quality control in the manager while keeping specialists single-purpose — a clean planner/worker split with built-in review. Because the manager allocates and validates at runtime, the crew adapts to tasks whose subtask breakdown isn’t known in advance, at the cost of extra manager LLM calls.

Source

CrewAI Documentation, Processes (sequential vs. hierarchical, manager_llm / manager_agent, Process.hierarchical): https://docs.crewai.com/en/concepts/processes

CrewAI Documentation, Hierarchical Process guide: https://docs.crewai.com/en/learn/hierarchical-process