# WHAT IS AUTOGEN? AND WHAT REPLACED IT > AutoGen came out of Microsoft Research and modelled work as a conversation between agents. It is now in maintenance mode, superseded by Microsoft Agent Framework — here is what it did, and what carries over. - Canonical: https://www.alpacamango.com/docs/what-is-autogen - Author: Alpaca Mango - Published: 2026-08-10 · Updated: 2026-08-10 - Topics: AI Agents, Agent Frameworks, Python **AutoGen is in maintenance mode.** Its README carries a maintenance-mode badge and states that it "will not receive new features or enhancements and is community managed going forward", and it directs new users to [Microsoft Agent Framework](https://github.com/microsoft/agent-framework) instead. If you are starting something today, start there. The rest of this page explains what AutoGen was, why its ideas still matter, and what carries over. ## What AutoGen did AutoGen modelled work as a *conversation between agents*. Rather than declaring a graph of steps, you defined participants and how they could talk, and the solution emerged from the exchange. Think of a group chat with rules: some participants are model-backed, some are code executors, one may be a human, and a speaker-selection policy decides who talks next. Its signature move — one agent writes a script, another runs it, and they loop on the traceback until the output is right — is worth stealing whatever you build on. Having a model write code, execute it and read the real error is dramatically more reliable than asking it to be correct first time, because it turns a language problem into an engineering one with a hard oracle attached. ## Why it was superseded The honest answer is that the conversational model is excellent at exploration and poor at constraint. A conversation that can go anywhere will occasionally go somewhere expensive, and "the agents talked past each other for forty turns" is a real failure mode with a real invoice. Cost and latency are hard to predict when the number of turns is emergent by design. Microsoft's own framing is that Agent Framework is the direct successor, created by the same teams, and that it *combines* AutoGen's simple agent abstractions with Semantic Kernel's enterprise features — session-based state management, type safety, middleware, telemetry — and adds graph-based workflows for explicit multi-agent orchestration. In other words: keep the abstractions, add the constraint. ## What to use instead [Microsoft Agent Framework](https://github.com/microsoft/agent-framework) (MAF) is an open, multi-language framework for production-grade agents and multi-agent workflows in .NET and Python — `agent-framework` on PyPI, `Microsoft.Agents.AI` on NuGet. Its workflow layer offers sequential, concurrent, handoff and group-collaboration patterns, with checkpointing, streaming, human-in-the-loop and time-travel; observability is OpenTelemetry out of the box. Microsoft publishes migration guides from both AutoGen and Semantic Kernel, which is the useful signal here: this is a consolidation of two of its own projects, not an abandonment. ## Should you migrate? Working software does not stop working because a badge changed. Maintenance mode means no new features and community management — not deletion. A running AutoGen system with tests around it is not an emergency. What you should not do is *start* something new on it, or plan a roadmap that assumes upstream will meet you halfway. If you are choosing today, choose the successor and read the migration guide before you have anything to migrate. ## Where to go next For the roles-and-tasks alternative, see [CrewAI vs Microsoft Agent Framework](/docs/crewai-vs-autogen). If what you actually need is explicit control over every step, the honest answer remains [LangGraph](/docs/langchain-vs-langgraph). For the wider map, see [AI agent frameworks compared](/docs/ai-agent-frameworks-compared).