Authority AI Editorial
CrewAI: role-based agent crews
CrewAI is a Python framework for building agents that work in named roles rather than as a single script or an explicit graph. It is aimed at developers who want autonomous collaboration between several agents, each with its own role and goal, alongside enough structured control to run the result in production. The reader this is for is someone who already runs coding agents and is deciding whether a role-based multi-agent model fits their next automation better than a single agent or a graph-based one.
What it does
CrewAI gives you two complementary building blocks. A Crew is a team of agents that collaborate autonomously: each agent gets a role, a goal, and a backstory that shapes how it approaches its assigned tasks, and the Crew runs a process, either sequential, where tasks execute one after another and each step can use the previous step's output, or hierarchical, where a manager agent delegates tasks to the others and reviews their output before moving on. A Flow is the other half: an event-driven workflow written as production Python code, giving you precise branching and state management around single LLM calls or entire Crews. The two are meant to be combined, using a Flow to hold the parts of an automation that need guaranteed control and a Crew for the parts where you want agents to reason and delegate on their own.
How you use it
CrewAI requires Python 3.10 through 3.13 and uses uv for dependency and package management; you install the CrewAI CLI with uv tool install crewai. Running crewai create crew <project_name> scaffolds a project with a JSON-first layout: agents live in individual .jsonc files, tasks and crew-level settings live in crew.jsonc, and crewai run loads that definition directly, prompting you for any placeholder inputs the crew's text references. An older Python and YAML scaffold, with crew.py plus config/agents.yaml and config/tasks.yaml, is still available with the --classic flag for anyone who prefers that pattern or is maintaining an existing project built on it. From there you connect each agent to a model provider by setting the llm field on the agent definition and supplying the matching API key.
What to know before adopting it
The core CrewAI framework is MIT licensed and free to use. CrewAI Inc. also sells a commercial layer called CrewAI AMP Suite on top of the open-source framework, adding managed deployment, tracing and observability, governance, and enterprise support, deployable on-premise or in the cloud; one piece of that suite, the Crew Control Plane, has a free tier at app.crewai.com. None of that is required to run the open-source framework yourself. The project maintains skill packages for coding agents, including a Claude Code plugin and a generic npx skills add crewaiinc/skills install for Cursor, Codex, and Windsurf, that teach an agent CrewAI's own scaffolding and configuration patterns, which is worth knowing if you are already running an agent that will be doing the CrewAI setup work for you.
Where it sits next to similar tools
CrewAI and LangGraph solve the same broad problem, coordinating multiple steps or agents toward one outcome, from different starting points. LangGraph asks you to model the whole flow as an explicit graph of nodes and edges; CrewAI asks you to describe agents by role and goal and let a process, sequential or hierarchical, decide the order, with Flows available when you need graph-like precision around specific parts. Compared to AutoGPT's current platform, which is a hosted or self-hosted product with a visual builder and marketplace, CrewAI stays a library you import into a codebase you own and version yourself, closer to how you would already be working if you write your agents in Python rather than assemble them visually. Anyone assembling instructions for an agent rather than choosing between these frameworks can also start from a fixed set rather than writing from scratch: the AI Ultimate Skill Collection bundles 3,029 ready-made skills in 12 packs, each a written procedure your agent follows step by step.