Skip to content

AI, decoded

Which AI agent framework should you use — LangGraph, CrewAI, or AutoGen?

LangGraph emphasizes explicit state and orchestration. CrewAI combines role-based teams with structured Flows. AutoGen supports multi-agent applications, but its README now says Microsoft has archived it and no longer maintains it, and directs new users to its successor, Microsoft Agent Framework. Choose around the workflow you need to control and the project’s current support status. CrewAI founder João Moura told Chain of Thought he hears the same thing across the market: no one wants lock-in, and teams want to change LLMs, frameworks, or memory at any point.

· Updated · Chain of Thought

Level 4: How AI systems are built · 4.3 Tools, MCP and agent architecture

AI AgentsMulti-Agent Systems

  • 2 episodes
  • 2 guest clips
  • 3 glossary terms

In detail

LangGraph: explicit orchestration

LangGraph’s documentation describes a low-level runtime for stateful agents. It supports durable execution, human intervention, and mixing deterministic steps with model-driven steps. It can be used without LangChain. Consider it when you need to make the workflow’s state and transitions explicit.

CrewAI: Crews and Flows

CrewAI’s documentation distinguishes Crews, which organize agents around roles and tasks, from Flows, which manage state and execution. Treating CrewAI as only a role-playing abstraction misses that second layer. Its documented production pattern starts with a Flow and delegates suitable work to a Crew.

AutoGen: check the migration path

AutoGen’s repository README now says the project is archived and no longer actively maintained by Microsoft, which is not providing bug fixes, security fixes, new features or support. The README advises against production use, describes AutoGen as community-managed, and limits contributions to bug fixes, security patches and documentation. It names Microsoft Agent Framework as AutoGen’s successor for new users and links a migration guide for existing users. That support status matters more to a new deployment decision than the older shorthand that AutoGen means agents chatting with one another.

Make the comparison on your own task

Our recommendation is to test a representative workflow before committing to an abstraction. Include a tool failure, a restart, a human approval, and a changed requirement. Compare how clearly each option lets you express and inspect those behaviors. A small application may need only a simple loop; more complex orchestration can justify a framework.

This is an editorial comparison of the linked documentation, checked October 24, 2026, not a performance benchmark. The related interviews provide historical context; they are not evidence of the projects’ current APIs or maintenance policies.

LangGraph, CrewAI and AutoGen: three different shapes of control Three panels, drawn from each project's primary documentation as checked on October 24, 2026. LangGraph: an explicit graph of steps with shared state, mixing deterministic steps and model-driven steps, with durable execution and a point where a human can intervene. CrewAI: a Flow that manages state and execution and delegates suitable work to a Crew of agents organized around roles and tasks. AutoGen: agents chatting with one another, dimmed behind an archived sign; Microsoft no longer maintains it and points new users to Microsoft Agent Framework, with a migration guide for existing users. Below, a track runs one representative workflow of your own through four events before you commit: a tool failure, a restart, a human approval and a changed requirement. This is an editorial comparison of documentation, not a benchmark. Three shapes of control, one status check What each project’s own docs say it is, checked October 24, 2026. Pick the shape your workflow needs. LangGraph explicit state and transitions shared state code model H Deterministic and model steps mixed, durable execution, a human (H) can step in. Usable without LangChain. CrewAI a Flow that delegates to a Crew Flow state + execution Crew Crews organize agents by roles and tasks. The documented production pattern starts with the Flow. AutoGen agents chatting, archived ARCHIVED no longer maintained New users are directed to Microsoft Agent Framework; a migration guide covers existing users. THEN RUN YOUR OWN WORKFLOW THROUGH EACH ONE a tool failure a restart a human approval a changed requirement Compare how clearly each lets you express and inspect those moments. A small app may need only a simple loop. Not a benchmark.
Pick the shape of control your workflow needs, and check support status first. An editorial reading of the linked docs, not a benchmark. Download the image

Hear it from the guest

“They wanna change LLMs at any point. They wanna change frameworks at any point. They wanna change, like, memory at any point. They would just wanna have flexibility.”
João Moura, CrewAI EP 29 · around 10:38 · read the transcript
“Agents, like as we start are usually kind of like straightforward, right? Like, Hey, I have an LLM. I have tools. I have a task. I would give you an outcome. What I find though, is that as you bring these things into production, especially like an enterprise company, then another, like a few other needs start to pop up.”
João Moura, CrewAI EP 2 · around 6:32 · read the transcript

Quotes lightly edited to remove filler words.

From the conversation

This explainer is drawn from these episodes — each carries its full transcript.

Concepts in this explainer

AI AgentAI BenchmarkLarge Language Model

Keep exploring