Drag each answer to line up with its matching prompt
CrewAI
Uses a GroupChat pattern where a moderator routes turns between agents in a conversation style multi-party loop
AutoGen
Assigns human like roles (Researcher, Writer, Reviewer) to LLM agents; coordination is through role based task delegation
LangGraph
Models the multi-agent workflow as a directed graph with typed state; coordination is through conditional edge transitions
CrewAI coordinates through role-based delegation, AutoGen through a moderated group chat, and LangGraph through an explicit state graph with conditional edges.
Imagine three ways to run a team project. In the first, you hand each person a job title and a task, and they pass work down the line. The researcher gathers facts, the writer drafts, the reviewer checks. That is CrewAI. In the second, everyone sits in one chat room and a facilitator decides who speaks next, so the work emerges from conversation. That is AutoGen. In the third, you draw a flowchart on the wall with arrows and decision diamonds, and the team follows the arrows, branching based on what they find at each box. That is LangGraph. All three get the same kind of work done, but the way control passes between people is completely different. The first leans on roles, the second on dialogue, and the third on an explicit map of who runs when.
Concept explanation~2 min read
Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example. Click to expand.
Concept explanation~2 min read
Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example. Click to expand.
Multi-agent frameworks exist because some tasks are better solved by several specialised agents than by one monolithic agent. But the moment you have more than one agent, you face a new question that single-agent loops never raise: how does control pass between agents, and who decides what runs next?
CrewAI, AutoGen, and LangGraph each answer that question differently. CrewAI answers with roles, AutoGen with conversation, and LangGraph with an explicit graph. The frameworks overlap heavily in what they can ultimately express, since each is Turing-complete enough to imitate the others with effort. So the real distinction is not capability but the coordination model each one makes natural and cheap to write.
That model is what an interviewer is probing when they ask you to match them. It determines how you debug a failure, how you add a branch, how you bound a run, and how easily you can certify the system for a regulated setting. Matching the framework to its model correctly signals that you understand multi-agent design as a control-flow decision rather than a library preference.
CrewAI: roles and task delegation
CrewAI models a multi-agent system as a crew of personas. You define each agent with a role, a goal, and a backstory, for example a Researcher whose goal is to gather sources or a Writer whose goal is to produce a draft. Those three fields are compiled into the agent's system prompt, so the persona is a prompt-level constraint rather than a weight-level one. You then declare tasks and assign them to agents, optionally with dependencies so one task's output feeds the next.
Coordination happens through delegation by role. In the sequential mode, tasks run in declared order and outputs flow down the chain. In the hierarchical mode, a manager agent decides which specialist to delegate each subtask to, and agents can also delegate to one another when their role calls for it. The mental model is staffing and running a small team, which makes CrewAI fast to express and easy to read for anyone who has ever organised people around responsibilities.
The tradeoff is that control flow is shallow. Because routing is driven by roles and task ordering rather than an explicit topology, complex branching, loops, and intermediate decision points become awkward to encode. A revision loop, where a reviewer can send work back, has no first-class representation and usually has to be faked through a manager agent or a custom callback. CrewAI shines when the work decomposes cleanly into specialist responsibilities and the path is mostly linear. It strains when the task needs many data-dependent forks that a flowchart would express more honestly.
Situations where this technique stops working.
2–4 min · Everything important, quickly.
Real products, models, and research that use this idea.
- A research-report pipeline in CrewAI uses a Researcher agent to gather sources, a Writer to draft, and a Reviewer to fact-check, with tasks chained in sequence.
- AutoGen powers code-generation assistants where a coder agent, an executor agent, and a critic agent debate fixes in a GroupChat until tests pass.
What an interviewer would ask next. Try answering before peeking at the approach.
QHow does adding a new conditional branch differ across CrewAI, AutoGen, and LangGraph?
In LangGraph you add a node and a conditional edge with a routing function. In CrewAI you add a task and adjust role delegation. In AutoGen you reshape the speaker-selection logic or prompt, since the branch lives in the conversation rather than an explicit topology.
Red flags & common mistakes
The phrases that signal junior thinking. Click to expand.
Red flags & common mistakes
The phrases that signal junior thinking. Click to expand.
Treating the three as interchangeable wrappers. They differ in their core coordination model, and that choice drives how you debug, branch, and control the workflow.
60 second bullets to scan on the way to the call.
State CrewAI's coordination model and name its persona-style roles.
State AutoGen's GroupChat pattern and how the next speaker is chosen.
Primary sources. Browse if you want the original framing.
Same topic, related formats. Practice these next.