AutoGen (Legacy / Migration)

• Key points are in the visual diagram above.

AutoGen solved conversational teams — a coder and reviewer discussing solutions. New greenfield work on the Microsoft stack should use Microsoft Agent Framework. This lesson is migration awareness: how AutoGen modeled agents as conversable entities that exchange messages.

AutoGen is like a group chat where each participant is an AI specialist — they discuss, debate, and build on each other's messages until the problem is solved.

Scenario

Coding task: UserProxy sends 'build a REST API for todos.' Coder agent writes code → UserProxy executes in Docker → Tester agent writes tests → Reviewer agent checks quality → iterate until tests pass.

Solution

In Microsoft Agent Framework, apply AutoGen (Legacy / Migration) to this scenario: Coding task: UserProxy sends 'build a REST API for todos. Identify the inputs, run the technique, validate the output, and note one thing you would monitor in production.

Do this before moving to the next module — reading alone is not enough.

Open the Code Walkthrough below and run it locally. Change one parameter related to AutoGen (Legacy / Migration) (e.g. model, temperature, top_k, or tool name), observe the difference in output, and write 2–3 sentences explaining what changed.

Highlighted lines show where AutoGen (Legacy / Migration) happens in the code.

1from autogen import AssistantAgent, UserProxyAgent, GroupChat, GroupChatManager  # import dependencies2
3coder = AssistantAgent(4    name="coder",5    system_message="You are a Python developer. Write clean, tested code.",6    llm_config={"model": "gpt-4o"},7)8
9reviewer = AssistantAgent(10    name="reviewer",11    system_message="You review code for bugs, security, and best practices.",12    llm_config={"model": "gpt-4o"},13)14
15user_proxy = UserProxyAgent(16    name="user",17    human_input_mode="NEVER",18    code_execution_config={"work_dir": "output", "use_docker": True},19)20
21group_chat = GroupChat(agents=[coder, reviewer, user_proxy], messages=[], max_round=10)22manager = GroupChatManager(groupchat=group_chat, llm_config={"model": "gpt-4o"})23
24user_proxy.initiate_chat(manager, message="Build a FastAPI todo app with tests")

• No max_round — agents converse indefinitely
• Code execution without Docker sandbox
• GroupChat for simple sequential tasks — overkill
• Not defining clear system messages per agent
• Ignoring AG2 migration for new projects

Quick recap — the most important points from this module.