Group Chat (Legacy)

• Key points are in the visual diagram above.

Two-agent ping-pong is easy. A researcher, coder, and critic need a manager or they talk forever.

A meeting with a facilitator and a talking stick. Without the stick, everyone talks over each other.

Scenario

Researcher → Coder → Critic, max_round=8, stop when Critic says APPROVE.

Solution

In Microsoft Agent Framework, apply Group Chat (Legacy) to this scenario: Researcher → Coder → Critic, max_round=8, stop when Critic says APPROVE. 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 Group Chat (Legacy) (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 Group Chat (Legacy) happens in the code.

1from autogen import AssistantAgent, GroupChat, GroupChatManager, UserProxyAgent  # import dependencies2
3researcher = AssistantAgent("researcher", llm_config=llm)4coder = AssistantAgent("coder", llm_config=llm)5user = UserProxyAgent("user", human_input_mode="NEVER")6chat = GroupChat(agents=[user, researcher, coder], max_round=8, speaker_selection_method="round_robin")  # key line for Group Chat (Legacy)7manager = GroupChatManager(groupchat=chat, llm_config=llm)  # key line for Group Chat (Legacy)8user.initiate_chat(manager, message="Write a pytest for fizzbuzz")  # key line for Group Chat (Legacy)

• No max_round
• Five agents with auto speak and no stop

Quick recap — the most important points from this module.