Multi-Agent Collaboration in CrewAI: Setup, Roles, and Goals

Build cooperative autonomous agents with defined responsibilities and shared objectives

Most LLM applications today are single-agent systems: one prompt in, one response out. But what if you could split responsibilities across a team of AI agents—just like a real organization? That’s what CrewAI enables. In this tutorial, you’ll learn how to create multi-agent workflows using CrewAI, assign roles and tools, and orchestrate shared goals with zero micromanagement.

🤖 Why Multi-Agent Design?

As tasks become more complex—like writing technical reports, performing research, analyzing sentiment, or even coding—it becomes inefficient (and error-prone) to rely on a single general-purpose agent.

Multi-agent collaboration brings structure by assigning:

• Roles: Each agent plays a specific part (e.g., Researcher, Editor, Planner)
• Goals: Agents have independent and shared objectives
• Tools: Agents operate with their own toolkits

CrewAI makes this pattern easy to implement without complex graph logic or orchestration engines.

Think of CrewAI as the AI version of a project team.

🛠 What You’ll Learn

By the end of this tutorial, you’ll:

• Define multiple AI agents with individual goals and distinct roles
• Assign tools and memory to each agent
• Orchestrate a multi-step task collaboratively (e.g., writing a product review)

🧰 Tools You’ll Use

🚀 Use Case: Product Review Assistant Team

Your AI team will:

1. Researcher Agent – Gathers information about a product
2. Writer Agent – Drafts a review based on research
3. Reviewer Agent – Checks grammar and logic, revises the content

✅ Step 1: Install Required Libraries

bashCopyEditpip install crewai langchain openai

✅ Step 2: Import CrewAI and LangChain Tools

pythonCopyEditfrom crewai import Agent, Task, Crew
from langchain.tools import DuckDuckGoSearchRun
from langchain.chat_models import ChatOpenAI

llm = ChatOpenAI(model_name="gpt-3.5-turbo", temperature=0)
search_tool = DuckDuckGoSearchRun()

✅ Step 3: Define Your Agents (Roles + Tools)

pythonCopyEditresearcher = Agent(
    role="Researcher",
    goal="Find technical specifications and customer reviews for the Apple Vision Pro",
    backstory="You're a tech analyst with expertise in XR hardware.",
    tools=[search_tool],
    llm=llm,
    verbose=True
)

writer = Agent(
    role="Writer",
    goal="Write a compelling product review based on the research",
    backstory="You're a persuasive tech journalist with excellent writing skills.",
    tools=[],
    llm=llm,
    verbose=True
)

reviewer = Agent(
    role="Reviewer",
    goal="Edit the review for clarity, tone, and grammar",
    backstory="You're a senior editor at a technology blog.",
    tools=[],
    llm=llm,
    verbose=True
)

✅ Step 4: Define Your Tasks

pythonCopyEdittask1 = Task(
    description="Gather technical specs and pros/cons of the Apple Vision Pro.",
    agent=researcher
)

task2 = Task(
    description="Write a 500-word review based on the research data.",
    agent=writer
)

task3 = Task(
    description="Review the draft and revise for tone and grammar.",
    agent=reviewer
)

✅ Step 5: Create the Crew and Kick Off

pythonCopyEditcrew = Crew(
    agents=[researcher, writer, reviewer],
    tasks=[task1, task2, task3],
    verbose=True
)

crew_result = crew.kickoff()
print(crew_result)

🔄 What Happens Behind the Scenes?

• CrewAI executes tasks in order based on the sequence you define.
• Each agent uses its goal, role, and backstory to guide how it responds.
• Agents can include tools (like search, calculators, vector DBs).
• Each output becomes context for the next task.

🧠 Pro Tips for Effective Collaboration

📚 Learn More / Expand This

• Add shared memory between agents using LangChain VectorStores
• Use CrewAI’s new support for agent feedback loops
• Define a Manager Agent to assign dynamic tasks (e.g., triaging incoming support tickets)

🔗 Useful Resources

• 📘 CrewAI Docs
• 🔍 DuckDuckGo Tool Wrapper
• 🧠 LangChain Agent Tools
• 🛠️ CrewAI GitHub
• 💬 CrewAI Community