AutoGen

AutoGen is an open-source framework from Microsoft Research for building applications powered by multiple AI agents that talk to each other to solve problems. Instead of a single prompt-response loop, you define several agents with different roles, such as one that writes code, one that reviews it, and one that executes it and reports results, and let them converse until the task is done. It supports human-in-the-loop checkpoints so a person can approve, redirect, or halt an agent conversation at any point.

The project underwent a major rewrite (AutoGen 0.4, later folded into Microsoft's AutoGen AgentChat and the broader Agent Framework work) that moved it to an asynchronous, event-driven architecture and split the codebase into layered components: a core runtime, AgentChat for common multi-agent patterns, and extensions for tools like web browsing or code execution. It's used by researchers experimenting with agent coordination and by developers building things like automated coding assistants, data analysis pipelines, and workflow orchestration where multiple specialized roles genuinely improve results over a single model call.

What sets it apart from other agent frameworks is Microsoft's backing and its early mover status: it was one of the first widely adopted multi-agent frameworks and helped popularize the pattern of agents debating, critiquing, and correcting each other's work. It competes with frameworks like CrewAI and LangGraph, and Microsoft has increasingly positioned it alongside Semantic Kernel as part of a unified agent development story.