Microsoft Agent Framework Python 2026: AutoGen's Successor Explained

Microsoft retired the AutoGen name and shipped Agent Framework. Here's what changed, what the Python API looks like, and when to use it.

💡 What You Will Learn

Microsoft retired the AutoGen name and shipped Agent Framework. Here's what changed, what the Python API looks like, and when to use it.

📜 Table of Contents

• The Succession Story
• What Agent Framework Gives You
• The Minimal Python Example
• AutoGen vs Agent Framework: The Decision
• Getting Started
• The Ecosystem Note
• FAQ

The Succession Story

Microsoft's AutoGen (60,404 stars, 2026-08-14) pioneered multi-agent conversations, but the project was split and reorganized in 2025: the core became Microsoft Agent Framework (12,774 stars), and AutoGen is now a lightweight, easy-to-learn layer on top of it. For new projects, start with Agent Framework; learn AutoGen only for its specific patterns.

What Agent Framework Gives You

• Agents as first-class objects - define agents, tools, and how they talk.
• Workflows - sequential, parallel, or network-based agent execution.
• Model-agnostic - OpenAI, Azure, local models, all through one interface.
• Observability - OpenTelemetry tracing built in.
• Python-first with a TypeScript counterpart.

The Minimal Python Example

from agent_framework import Agent, Runner
from agent_framework.tools import function_tool

@function_tool
def get_weather(city: str) -> str:
    return f"sunny, 24C in {city}"  # real API call in production

agent = Agent(
    name="WeatherBot",
    instructions="Answer using the weather tool when asked about weather.",
    tools=[get_weather],
)

result = Runner.run_sync(agent, "What's the weather in Tokyo?")
print(result.final_output)

AutoGen vs Agent Framework: The Decision

Getting Started

pip install agent-framework

Then follow the official quickstart: define an agent with a tool, run it, add OpenTelemetry tracing. The docs are clean and the examples are runnable.

The Ecosystem Note

Microsoft also maintains Semantic Kernel (28,445 stars) - the SDK for integrating LLMs into existing apps (plugins, planners, memory). Rule of thumb: Semantic Kernel for 'add AI to my app', Agent Framework for 'build a multi-agent system'.

FAQ

Is AutoGen dead? No - it lives on as a simplified layer; the production core moved to Agent Framework.

Is it free? Open source (MIT); you pay only for model usage.

Python or TypeScript? Both are supported; Python has the richest examples.

How is it different from LangChain? LangChain is a broad integration library; Agent Framework is Microsoft's opinionated agent runtime with workflows built in.