CrewAI — Module 14.3

How This Lesson Fits the Module & Volume

RAG chains answer questions; multi-agent systems divide labor. CrewAI packages that idea as crews of role-playing agents with tasks and tools—a practical preview of Volume 15 multi-agent systems.

Compare with AutoGen (conversation-centric) and LangGraph (explicit graphs) when choosing a multi-agent style.

Learning Objectives

By the end of this lesson, students should be able to:

Define CrewAI agents, tasks, tools, and crews.
Sketch a researcher + writer crew for a content workflow.
Contrast role-based crews with single RAG chains.
Identify failure modes: loops, cost blowups, vague roles.
Place CrewAI among AutoGen and LangGraph options.
Link crew design to Volume 15 multi-agent concepts.

Definition

CrewAI is a Python framework for orchestrating role-based AI agents that collaborate as a crew. Each agent has a role, goal, and backstory; tasks define expected outputs; processes (sequential/hierarchical) coordinate how work flows.

Multi-Agent vs RAG Orchestration

Dimension	RAG chain	CrewAI crew
Primary job	Grounded Q&A	Delegated multi-step work
Actors	One pipeline	Multiple role agents
Control	Fixed steps	Process + task graph
Cost profile	Predictable-ish	Can explode with chatter
Best fit	Knowledge answers	Research, drafting, review

Minimal Crew Sketch

from crewai import Agent, Task, Crew, Process

researcher = Agent(
    role="Research Analyst",
    goal="Find accurate facts about the topic",
    backstory="Careful analyst who cites sources",
    verbose=True,
)

writer = Agent(
    role="Technical Writer",
    goal="Turn research into a clear brief",
    backstory="Explains complex topics simply",
    verbose=True,
)

research_task = Task(
    description="Research enterprise SSO options for mid-size SaaS.",
    expected_output="Bullet list of options with pros/cons and sources.",
    agent=researcher,
)

write_task = Task(
    description="Write a one-page decision brief from the research.",
    expected_output="Markdown brief under 400 words.",
    agent=writer,
)

crew = Crew(
    agents=[researcher, writer],
    tasks=[research_task, write_task],
    process=Process.sequential,
)
result = crew.kickoff()
print(result)

Strengths

Intuitive role metaphors
Fast multi-agent demos
Task/output contracts
Good teaching bridge to Vol 15

Tradeoffs

Costly multi-LLM chatter
Harder to constrain than graphs
Role overlap causes thrash
Still need eval & guardrails

Common Misconception

“More agents = better results.” Extra agents multiply tokens and coordination failures. Start with two clear roles and measurable task outputs; add agents only when a distinct specialty is proven.

Knowledge Check

Short Answer: What is a CrewAI crew? Answer: A coordinated group of role-based agents with tasks/process.
True/False: Agents typically have role, goal, and backstory. Answer: True.
Multiple Choice: Sequential process means: (a) random talk, (b) tasks run in order, (c) only FAISS search. Answer: (b).
Short Answer: Name a good use case for crews vs RAG. Answer: Multi-step research/writing/review workflows (any valid).
True/False: Adding agents never increases cost. Answer: False.
Multiple Choice: AutoGen is relatively more: (a) conversation-oriented, (b) SQL-only, (c) CNN training. Answer: (a).
Short Answer: Why define expected_output on tasks? Answer: To constrain deliverables and improve reliability.
Short Answer: Which Volume covers multi-agent systems in depth? Answer: Volume 15.
Multiple Choice: Vague overlapping roles often cause: (a) perfect determinism, (b) thrash/duplication, (c) free GPUs. Answer: (b).
True/False: CrewAI replaces vector databases for RAG facts. Answer: False—crews may still use tools/RAG underneath.

Key Takeaways

CrewAI models collaboration as role agents + tasks + process.
Use crews for delegated work; use RAG chains for grounded Q&A.
Control cost with few roles and crisp expected outputs.
Compare with AutoGen conversations and LangGraph state machines.
Next: AutoGen for multi-agent conversation patterns.

Trainer’s Guide

Role play: Students write role/goal/backstory cards before coding; peers spot overlaps.

Lab: Two-agent crew with a mock search tool; log token usage per kickoff.

Recap: CrewAI turns multi-agent ideas into role-based crews. Continue with AutoGen.