Building with AI

CrewAI

4 min readintermediateUpdated 28 Sept 2026
1 · In one line

CrewAI is an open-source toolkit, written in Python, for building teams of AI agents, each with a role, that work through a list of tasks together.

1 · What it is

Some jobs have several parts that need different skills, such as finding facts and then writing them up. CrewAI is an open-source toolkit, written in Python, for this. You build a small team of AI agents, called a crew, and give it a list of tasks.

Each agent gets a role, a goal and a short backstory. These tell the agent what it is for. An agent can also be given tools, which are extra abilities the agent can call on. Each task has a description, the output it should produce and the agent responsible. Picture a crew with a researcher and a writer. The researcher finds facts about a topic, and the writer turns them into an article.

The crew follows a process. The default is sequential: tasks run in list order. Each result is passed on as background for the next task. In the hierarchical process, a manager agent is in charge. It plans the work, assigns tasks to agents based on what they can do and checks the results. You must give it a manager model or a manager agent.

For engineers: CrewAI also has Flows. A flow is a workflow written in ordinary code. Each step runs when something it waits for happens (that is what “event-driven” means). A flow remembers its data between steps, which is called keeping state, and it can branch or loop. It can hand a hard step to a crew and carry on with the result. Crews give flexible teamwork; flows give precise control. By default, agents cannot pass work to each other; that is called delegation, and it is off. CrewAI can work with many different language models, even ones that run on your own computer.

2 · Why it exists

One general AI helper struggles with a big job that has several different parts.

Too many hatsA single agent has to research, write and check, with no clear focus for each part.
No running orderWithout a plan, it is unclear which piece of work comes first and who receives the result.
Loose controlSome parts of an app need fixed steps and saved data, not free-form agent choices.
3 · How it works

Follow one article through a two-agent crew.

In the default sequential process, each task's output becomes context for the next task.
  1. 1 · defineYou describe each agent with a role, a goal and a backstory, and can give it tools.
  2. 2 · assignYou write each task with a description, the expected output and the agent responsible.
  3. 3 · groupYou put the agents and tasks into a crew and choose a process.
  4. 4 · passIn the sequential process, tasks run in list order and each output is context for the next.
  5. 5 · steerA flow can wrap crews with fixed steps, saved state and branching logic.

A crew is a team with a running order: roles, tasks and a process.

4 · Where it's used
WhoWhat they askWhat it works with
Marketing team“Can one agent research a topic and another write the post?”A sequential crew of a researcher and a writer
Operations team“Can a manager agent decide who handles each request?”A hierarchical crew with a manager agent
Product team“Can we run fixed steps and call a crew only for the hard part?”A flow that starts a crew and keeps state
Data team“Can agents use extra tools as they work?”Tools given to each agent
5 · What it solves, and what it doesn't
solves
  • It lets you split a job across agents that each have a clear role and goal.
  • It passes one task's output to the next task as context.
  • It offers a manager option that assigns tasks and reviews results.
  • Flows add fixed steps, saved state and branching around the agents.
doesn't solve
  • It does not choose your roles or tasks; you still design the team.
  • Crews let agents decide for themselves; for precise, step-by-step control you need a flow instead.
  • The hierarchical process needs a manager model or manager agent that you set.
  • Each agent still depends on the language model behind it.
6 · Go deeper

Sources used

This explainer is written in original language. The links below support its factual claims.

  1. repocrewAIInc/crewAI, CrewAI (GitHub) · read 28 Sept 2026
  2. docsIntroduction, CrewAI documentation · read 28 Sept 2026
  3. docsAgents, CrewAI documentation · read 28 Sept 2026
  4. docsTasks, CrewAI documentation · read 28 Sept 2026
  5. docsCrews, CrewAI documentation · read 28 Sept 2026
  6. docsProcesses, CrewAI documentation · read 28 Sept 2026
  7. docsFlows, CrewAI documentation · read 28 Sept 2026