CrewAI: Multi-Agent Teams with Roles and Tasks

After this lesson, you will be able to: Define specialised agent roles, assign tasks, and orchestrate a CrewAI crew that collaborates to complete a research workflow.

CrewAI is the Python answer to multi-agent teams. You define agents (with roles, goals, backstories), tasks (with descriptions and expected outputs), and a crew that runs them. This lesson builds a 3-agent research crew that researches, drafts, and reviews, fully in code.

Prerequisites:LlamaIndex: RAG Pipelines and Data Agents

Install + setup

1
1. `pip install crewai crewai-tools`
2
2. Set ANTHROPIC_API_KEY (or OPENAI_API_KEY).

A 3-agent research crew

Researcher → Writer → Reviewer.

python

from crewai import Agent, Task, Crew, Process
from crewai.llm import LLM
from crewai_tools import SerperDevTool

llm = LLM(model='anthropic/claude-sonnet-4-6')
search_tool = SerperDevTool()  # needs SERPER_API_KEY (free at serper.dev)

researcher = Agent(
    role='Senior Research Analyst',
    goal='Find accurate, current facts about a topic',
    backstory='You are obsessive about citations and primary sources.',
    tools=[search_tool], llm=llm, verbose=True,
)
writer = Agent(
    role='Tech Writer',
    goal='Turn research into a clear 400-word piece',
    backstory='You write for smart non-experts.',
    llm=llm, verbose=True,
)
reviewer = Agent(
    role='Editor',
    goal='Catch factual errors, vague claims, and tone issues',
    backstory='Brutal but fair.',
    llm=llm, verbose=True,
)

research_task = Task(description='Research: {topic}', expected_output='5 cited bullet points', agent=researcher)
write_task = Task(description='Draft a 400-word post.', expected_output='Markdown post', agent=writer, context=[research_task])
review_task = Task(description='Review the draft, return the final.', expected_output='Polished markdown', agent=reviewer, context=[write_task])

crew = Crew(agents=[researcher, writer, reviewer], tasks=[research_task, write_task, review_task], process=Process.sequential, verbose=True)
result = crew.kickoff(inputs={'topic': 'state of AI agent frameworks in 2026'})
print(result)

When CrewAI shines

When the workflow is naturally team-shaped (research-write-review, plan-build-test, scout-analyze-decide). When prompts get unwieldy in a single agent. When you want auditability per role.

When NOT to use CrewAI

Single-step tasks. Latency-sensitive flows (each agent = 1+ LLM calls = compounding latency). Strict cost budgets. Use a single tightly-prompted agent with tools instead.

←LlamaIndex: RAG Pipelines and Data Agents

Back to Code-First Agents

Raw SDK: Anthropic and OpenAI APIs from Scratch→