AutoGen: Microsoft Research's multi-agent framework, now archived

The status you need to know first

As of October 22, 2026, the official AutoGen GitHub repository carries a clear warning: the project is archived and no longer maintained by Microsoft. No security fixes, no new features, no support. Microsoft notes that AutoGen was built for research and experimentation and advises against production use, especially with sensitive data. The last release dates back to September 2025.

New users are asked to start with Microsoft Agent Framework, which Microsoft’s documentation calls the “next generation” of both AutoGen and Semantic Kernel, built by the same teams.

What AutoGen brought to the table

Born at Microsoft Research in 2023, AutoGen spread a simple idea: instead of one AI agent, several agents that talk to each other to make progress. One agent writes code, another runs it, a third critiques the result, and a human can step in at any point. It was one of the projects that made the multi-agent system tangible for developers.

The framework was later fully re-architected (version 0.4) into layers: AgentChat for common use cases, Core for event-driven architectures, and Extensions for models and tools, including MCP support. It came in Python and .NET, and shipped AutoGen Studio, a no-code graphical interface for prototyping agent teams.

If you still use it

In practice you have three options:

1. Migrate to Microsoft Agent Framework using the official “from AutoGen” guide. Agents and multi-agent orchestration carry over, and you gain explicit workflows, state management and telemetry.
2. Freeze your version for an internal prototype, knowing that any vulnerabilities will not be patched.
3. Switch ecosystems if you are not tied to Microsoft.

Be careful not to confuse AutoGen with AG2, a community fork that is independent of Microsoft.

Alternatives

Outside the Microsoft world, CrewAI builds on the idea of agents collaborating through roles and is quick to pick up.