AutoGen was one of the earliest widely-used multi-agent frameworks, built by Microsoft Research to let multiple LLM agents converse with each other and with humans to solve tasks collaboratively. It was influential early on, shaping how a lot of developers first thought about multi-agent design patterns, even for people who ended up building on other frameworks later.
Microsoft put it into maintenance mode in October 2025, meaning the core team has stopped active feature development and the project now runs mostly on bug fixes and documentation updates from community contributors. Microsoft has pointed new production projects toward its successor, Microsoft Agent Framework, which suggests AutoGen's role going forward is mostly historical and educational rather than as a foundation for new work.
It's still genuinely useful for prototyping and for teaching multi-agent concepts, since its codebase and documentation are mature and well understood. But anyone starting a new production system today has reason to look at Microsoft Agent Framework, or at a more actively developed competitor, instead.