AutoGen, AG2, or Microsoft Agent Framework: Which One Should You Use in 2026?

Microsoft's agent ecosystem split into four branches. Here is the plain-English guide to what each one is and which one to build on.

The Situation

If you have searched for AutoGen tutorials recently, you have probably noticed something confusing: some tutorials use 'autogen', some use 'ag2', some reference 'pyautogen', and some mention 'Microsoft Agent Framework'. They install differently, have different APIs, and the docs do not clearly explain the relationship between them.

This is not your confusion -- it is a genuine fragmentation in Microsoft's agent ecosystem. Here is the full picture.

The History in Plain English

AutoGen started as a Microsoft Research project. It became popular quickly (tens of thousands of GitHub stars) for its multi-agent group chat pattern. Then:

1. Microsoft rewrote AutoGen from scratch as v0.4. The new version had a completely different, async-first, event-driven architecture. It was incompatible with v0.2.
2. The community that had built on v0.2 did not want to throw away their code. A team forked v0.2 and continued it independently as AG2.
3. Microsoft continued developing AutoGen v0.4 under the autogen-agentchat package.
4. In October 2025, Microsoft merged AutoGen v0.4 and Semantic Kernel into a unified product called Microsoft Agent Framework -- the new flagship for production enterprise deployments.

AutoGen v0.2 (pyautogen) and Semantic Kernel are now in maintenance mode. Microsoft will fix bugs and security issues but will add no new features. If you are starting a new project, do not build on these.

The Four Options Today

When to Use Each

AG2 -- if you want the AutoGen v0.2 experience, actively maintained

AG2 is the direct continuation of AutoGen v0.2 with the same group chat API, the same agent patterns, and an actively growing community. If you liked AutoGen v0.2 and do not need Azure enterprise features, AG2 is the path forward.

pip install ag2

# AG2 -- API is nearly identical to AutoGen v0.2
import ag2
 
assistant = ag2.AssistantAgent(
    name="assistant",
    llm_config={"model": "gpt-4o", "api_key": "..."},
)
 
user = ag2.UserProxyAgent(
    name="user",
    human_input_mode="NEVER",
    max_consecutive_auto_reply=5,
)
 
user.initiate_chat(assistant, message="Summarise the latest AI news.")

Microsoft Agent Framework -- if you need enterprise production features

The Microsoft Agent Framework is the production-grade successor that combines AutoGen's multi-agent orchestration with Semantic Kernel's plugin and memory system. It adds Azure Active Directory identity, RBAC, SOC 2 compliance, and multi-language support (Python, C#, Java).

pip install microsoft-agents  # preview as of early 2026

This is the right choice if your organisation needs audit logs, enterprise SSO, compliance certification, or is already invested in the Azure ecosystem.

AutoGen v0.4 / AgentChat -- transitional

AutoGen v0.4 (autogen-agentchat) is the bridge between the old AutoGen and the Microsoft Agent Framework. If you are migrating from v0.2, v0.4 is the intermediate step. For new projects, go directly to Microsoft Agent Framework or AG2.

Migration Paths

# Migrating from pyautogen to AG2:
pip uninstall pyautogen
pip install ag2
 
# Most imports change from:
# import autogen
# to:
# import ag2
# The core Agent classes, group chat, and LLM config patterns are compatible.

Quick Reference

• AutoGen v0.2 (pyautogen): maintenance mode -- do not start new projects here
• AG2: active community fork of v0.2, same API, good for non-Azure production use
• AutoGen v0.4 (agentchat): transitional -- folded into Microsoft Agent Framework
• Microsoft Agent Framework: Microsoft's new flagship, enterprise features, Azure-native
• Semantic Kernel: also in maintenance mode, features merged into Agent Framework
• For new projects: AG2 (community, any cloud) or Microsoft Agent Framework (enterprise, Azure)