LangChain vs CrewAI vs AutoGen 2026: AutoGen Is Archived

Most comparisons of these three still treat AutoGen as a peer. It is not any more, so this guide starts with what happened to it and then makes the comparison people searching this question actually need: LangGraph, CrewAI and AutoGen's successor, Microsoft Agent Framework (MAF). Price does not separate them, because all three are free. What separates them is how much control you want over flow and state versus how fast you want a working multi-agent system. For the wider field including the no-framework path, see the solo-founder roundup.

1. What happened to AutoGen

AutoGen's GitHub README tells the story^[3]:

• October 2025: the README told new users to start with Microsoft Agent Framework instead, while promising that AutoGen would keep receiving bug fixes and critical security patches.
• April 2, 2026: Microsoft Agent Framework 1.0 shipped for Python and .NET^[5].
• October 16, 2026: the README was changed to "Project archived and no longer maintained": no bug fixes, security fixes, new features or support, and a recommendation against production use.

The code is still on GitHub and still installs, so an existing AutoGen app does not stop working. It just stops getting fixed, including for vulnerabilities in its dependencies. Microsoft publishes a migration guide from AutoGen to Agent Framework^[6]; if you run AutoGen in anything customer-facing, that migration belongs on your roadmap now.

2. The control-vs-convenience axis

The three live options line up along one spectrum: how much of the orchestration you author by hand versus how much the framework decides for you.

Framework	Mental model	Position on the axis
LangGraph[1]	Graph of nodes and edges	Most control; you wire flow and state explicitly
Microsoft Agent Framework[4]	Agents plus graph-based workflows	Explicit workflows with ready-made patterns: sequential, concurrent, handoff, group collaboration
CrewAI[2]	Roles and crews	Fastest start; the framework handles coordination

LangGraph asks you to draw the state machine, which is more upfront work but leaves nothing implicit when a flow misbehaves. CrewAI lets you describe roles and a goal and assembles the coordination for you, so the first working version arrives fast but the orchestration is more of a black box. Microsoft Agent Framework sits between them: its workflows are graphs you define, with checkpointing, streaming and human-in-the-loop steps built in, and its orchestration patterns cover the common multi-agent shapes without wiring each one yourself^[4]. If you liked AutoGen's style of agents that message each other, MAF's group-collaboration pattern is the closest equivalent.

3. Which one a solo founder maintains best

For a one-person team the deciding question is not which builds fastest on day one, but which you can debug alone at 2 a.m. in month six:

• LangGraph costs more to learn, but when an agent loops or stalls, the graph shows exactly where, so the failure is inspectable. That pays off as the workflow grows complex.
• CrewAI gets a prototype running quickest, which suits validating an idea, but the abstracted coordination can be harder to reason about once behavior gets subtle.
• Microsoft Agent Framework is the youngest of the three at 1.0 (April 2026) and is built for production: OpenTelemetry tracing is built in, and it ships Python and .NET SDKs with consistent APIs^[4][5]. It is the natural pick if your stack is .NET or you deploy on Microsoft's cloud.

So the honest solo-maintainability read is: reach for CrewAI to ship a first version fast, LangGraph when the flow becomes stateful and branchy enough that you need to see every transition, and Agent Framework when you want those workflows with Microsoft's long-term support behind them or work in .NET. Prototype the same small task in your top two before committing.

4. Price is a wash; the paid layers are not

All three frameworks are free and open source (LangGraph and Agent Framework under MIT), so two systems calling the same model with the same prompts cost the same in tokens regardless of which you picked^[1][2][4]. Your bill is LLM tokens plus a small server, both driven by model and prompt design rather than the framework. Model that spend with the AI stack cost calculator and check it against a vendor reprice with the model price drop stress test.

Where money does differ is the optional managed tiers, which you pay only if you opt in:

• LangChain (LangSmith with Deployment): Developer at $0 per seat with up to 5,000 base traces a month; Plus at $39 per seat per month with up to 10,000 base traces, access to Deployment and pay-as-you-go usage beyond^[7].
• CrewAI AMP: Basic at $0 per month with 50 workflow executions and a visual editor; Enterprise custom^[8].
• Microsoft Agent Framework: agents can be deployed to Microsoft Foundry's hosted infrastructure, a paid cloud service billed separately, or run on any server you choose^[4].

5. Which to choose

• Complex stateful flows, explicit control: LangGraph.
• Role-based crews, fast to reason about: CrewAI.
• Production workflows with built-in multi-agent patterns, or a .NET stack: Microsoft Agent Framework.
• Already on AutoGen: keep it running, plan the move to Agent Framework with Microsoft's migration guide, and do not start new work on it^[6].
• Lowest cost on any of them: self-host, pay only tokens and a small server.

Re-verify the optional platform prices before relying on them; managed tiers reprice often, but the open-source frameworks remain free. For a solo-founder-focused shortlist, see the best AI agent frameworks for solo founders.

Framework status checked on GitHub on 2026-10-19; pricing figures checked against the official pricing pages on 2026-10-18.

Frequently asked questions

Is AutoGen still maintained?

No. Microsoft's AutoGen README now says the project is archived and no longer maintained: no bug fixes, security fixes, new features or support, and Microsoft advises against production use. Since October 2025 the README had pointed new users to its successor, Microsoft Agent Framework, which reached 1.0 for Python and .NET on April 2, 2026; the October 16, 2026 update withdrew the promise of security fixes. The code stays on GitHub and keeps working, but new projects should start on Agent Framework, LangGraph or CrewAI, and existing AutoGen apps should plan a migration using Microsoft's migration guide.

Is LangChain, CrewAI, or Microsoft Agent Framework free?

Yes. LangChain and its agent runtime LangGraph are MIT-licensed, CrewAI's orchestration framework is open source, and Microsoft Agent Framework is MIT-licensed, all with no licensing fee. What costs money is what you run them against: the LLM API tokens your agents consume, the compute you host them on, and any optional managed platform such as LangSmith Deployment, CrewAI AMP or Microsoft Foundry's hosted agents. The framework choice does not change your token bill; your model and prompt design do.

What is the real cost of running an agent framework?

Three things, none of which is the framework license. First and usually largest is LLM API tokens: a multi-agent system can make many model calls per task, so token spend dominates. Second is hosting: a small server or serverless functions to run the agents. Third is optional tooling like observability or a managed deployment platform. For a solo founder, the practical lever is reducing token spend through model choice, caching, and limiting agent loops, not picking a different open-source framework. Model your token usage before assuming the framework matters to cost.

How much does CrewAI's managed platform cost?

CrewAI's framework is free and open source, while its managed AMP platform has a Basic plan at $0 per month that includes 50 workflow executions, a visual editor, and GitHub integration, with Enterprise pricing custom, per CrewAI's pricing page on October 18, 2026. So you can build and run CrewAI entirely for free by self-hosting the open-source framework and paying only for LLM tokens, and the paid platform is an optional convenience layer for teams that want the hosted studio and enterprise controls rather than a requirement to use CrewAI at all.

Does LangGraph Platform cost extra on top of LangChain?

Yes, if you use it. LangChain and LangGraph are free open source, but managed deployment is now part of LangChain's LangSmith plans: Developer is $0 per seat with up to 5,000 base traces a month, and Plus is $39 per seat per month with up to 10,000 base traces and access to Deployment, pay-as-you-go beyond, per LangChain's pricing page on October 18, 2026. You do not need LangGraph Platform to use LangGraph; you can self-host the open-source runtime and skip the platform fees entirely, paying only for LLM tokens and your own infrastructure.