Group chat ending prematurely | AG2 (formerly AutoGen) | Page 1

For some reason after just 5-6 messages the group chat ends even though agents are specified to not exit until the task is complete.

The model used is gpt-3.5-turbo and the chat has 4 agents that should help each other, many of my tests end even before some of them talk.

# Create user proxy agent, coder, product manager
user_proxy = autogen.UserProxyAgent(
    name="User_proxy",
    system_message="A human admin who will give the idea and run the code provided by Coder.",
    code_execution_config={"last_n_messages": 1, "work_dir": "groupchat"},
    human_input_mode="ALWAYS",
)
coder = autogen.AssistantAgent(
    name="Coder",
    system_message="You will develop the entire tool code and won't leave until it is complete and working accordingly, if you need low level help ask Coder2, if you need high level help ask Coder Master. NEVER LEAVE ANYTHING TO BE DONE, TODO IS NOT ALLOWED YOU MUST CODE AND IMPLEMENT EVERYTHING THAT IS NEEDED.",
    llm_config=llm_config,
)
coder2 = autogen.AssistantAgent(
    name="Coder2",
    system_message="You will help Coder with common stuff that he may struggle with, if you both get stuck, you will ask for help from Coder_master.",
    llm_config=llm_config,
)
coder_master = autogen.AssistantAgent(
    name="Coder_master",
    system_message="You will help Coder to structure the tool, decide on the most optimal way to fix issues and continue checking that everything in the plan gets implemented before exiting the chat.",
    llm_config=llm_config,
)
pm = autogen.AssistantAgent(
    name="product_manager",
    system_message="You will help break down the initial idea into a well scoped requirement for the coder; Do not involve in future conversations or error fixing",
    llm_config=llm_config,
)

The chat is initialized in the following way:

# Create groupchat
groupchat = autogen.GroupChat(
    agents=[user_proxy, coder, coder2, coder_master, pm], messages=[])
manager = autogen.GroupChatManager(groupchat=groupchat, llm_config=llm_config)

# Start the conversation
user_proxy.initiate_chat(
    manager, message="my specific programming task")

#Group chat ending prematurely