AutoGen still orchestrates agents, but Microsoft has moved on

What it does

AutoGen is a layered Python and .NET framework for building applications where multiple LLM agents pass messages, call tools, and coordinate on tasks. It offers the low-level Core API for message passing and distributed runtimes, the higher-level AgentChat API for rapid prototyping, and an Extensions API for plugging in model clients and external tools like MCP servers. Microsoft also provides AutoGen Studio, a no-code interface for prototyping workflows, though the README explicitly warns it is not meant for production.

The interesting bit

The README reads less like a product pitch and more like a historical plaque: a 59,278-star repository that pioneered experimental multi-agent orchestration is now frozen in maintenance mode while Microsoft redirects all new development to its enterprise successor, the Microsoft Agent Framework.

Key highlights

• Three-layer architecture: Core handles message passing and local or distributed runtimes with cross-language support; AgentChat provides an opinionated API for common patterns like group chats; Extensions adds LLM clients and capabilities such as code execution.
• Agents can integrate with MCP servers and compose other agents via AgentTool, enabling setups where a general assistant delegates to specialized experts.
• AutoGen Studio offers a visual, no-code environment for building multi-agent workflows.
• The project is community-managed and ships with a benchmarking suite, AutoGen Bench.

Caveats

• AutoGen is in maintenance mode and will not receive new features or enhancements.
• AutoGen Studio is explicitly described as a prototyping and demonstration tool, not a production-ready application.
• Community-managed support means responses may be limited.

Verdict

Existing AutoGen users should consult the migration guide to Microsoft Agent Framework; newcomers should start there instead. The codebase remains a useful reference for layered agent architecture, but treat it as legacy.