CrewAI — Multi-Agent Orchestration in Python

What it is

CrewAI is a Python framework for building multi-agent systems where each agent has a defined role, goal, and set of tools. Agents collaborate on complex tasks through a crew structure that manages delegation, task ordering, and result synthesis. The framework supports sequential and hierarchical process types.

CrewAI targets developers building AI systems that require multiple specialized agents working together: research teams, content pipelines, data analysis workflows, or any task that benefits from division of labor among AI agents.

How it saves time or tokens

This workflow provides pip install and a working crew definition. Instead of building agent coordination logic from scratch, you define agents declaratively and let CrewAI handle task delegation, context passing, and result aggregation. The role-based model makes it intuitive to decompose complex tasks.

How to use

1. Install CrewAI:

pip install crewai crewai-tools

1. Define agents and tasks:

from crewai import Agent, Task, Crew

researcher = Agent(
    role='Senior Researcher',
    goal='Find the latest AI trends',
    backstory='Expert in AI research with 10 years of experience',
    tools=[search_tool]
)

writer = Agent(
    role='Technical Writer',
    goal='Create clear, engaging content',
    backstory='Experienced tech writer specializing in AI topics'
)

research_task = Task(
    description='Research the top 5 AI trends for 2026',
    agent=researcher,
    expected_output='Detailed report with sources'
)

writing_task = Task(
    description='Write a blog post based on the research',
    agent=writer,
    expected_output='1000-word blog post'
)

crew = Crew(
    agents=[researcher, writer],
    tasks=[research_task, writing_task]
)

result = crew.kickoff()

1. Run the crew and collect results.

Example

from crewai import Agent, Task, Crew, Process

# Hierarchical process with a manager
analyst = Agent(role='Data Analyst', goal='Analyze data patterns')
strategist = Agent(role='Strategy Consultant', goal='Develop recommendations')

crew = Crew(
    agents=[analyst, strategist],
    tasks=[
        Task(description='Analyze Q1 sales data', agent=analyst),
        Task(description='Create growth strategy', agent=strategist)
    ],
    process=Process.hierarchical,
    manager_llm='gpt-4o'
)

result = crew.kickoff()
print(result)

Related on TokRepo

• CrewAI multi-agent page -- Deep dive into CrewAI capabilities
• Multi-agent frameworks -- Compare agent orchestration tools

Common pitfalls

• Each agent makes separate LLM API calls, so a crew with 5 agents on a 3-task pipeline can make 15+ API calls. Monitor costs when using expensive models.
• Agent backstories significantly influence output quality. Vague backstories produce generic results; specific backstories produce focused outputs.
• Hierarchical process requires a manager LLM, which adds another API call per task delegation. Use sequential process for simpler workflows.