Group Chat Auto Pattern

You are creating an AG2 group chat workflow using AutoPattern -- fully LLM-driven agent selection with no handoffs.

Instructions

1. Ask the user for:

   • What task the group needs to solve
   • How many agents and their specializations
   • Maximum conversation rounds (default: 15)

2. Create the group chat following this pattern:

AutoPattern Group Chat

from autogen import ConversableAgent, UserProxyAgent, LLMConfig
from autogen.agentchat import run_group_chat
from autogen.agentchat.group.patterns import AutoPattern

llm_config = LLMConfig({"api_type": "anthropic", "model": "claude-sonnet-4-6"})

# Each agent MUST have a description -- used by Group Chat Manager for routing
agent_a = ConversableAgent(
    name="agent_a",
    system_message="Your role instructions here...",
    description="When to select this agent -- used for routing decisions.",
    llm_config=llm_config,
)

agent_b = ConversableAgent(
    name="agent_b",
    system_message="Your role instructions here...",
    description="When to select this agent -- used for routing decisions.",
    llm_config=llm_config,
)

user = UserProxyAgent(
    name="user",
    code_execution_config=False,
)

# AutoPattern -- no handoffs, LLM picks next agent based on descriptions
pattern = AutoPattern(
    initial_agent=agent_a,
    agents=[agent_a, agent_b],
    group_manager_args={"llm_config": llm_config},
    user_agent=user,
)

result = run_group_chat(
    pattern=pattern,
    messages="Your task here",
    max_rounds=15,
)
result.process()
print(result.summary)

Key Rules

• AutoPattern requires no handoffs -- the Group Chat Manager decides routing based on agent description fields
• Every agent MUST have a distinct description (not just system_message) -- this is what the Group Chat Manager uses for selection
• The system_message tells the agent how to behave; the description tells the manager when to select the agent
• Use LLMConfig({...}) -- NOT a raw dict like {"model": "..."}
• Use run_group_chat with a pattern -- NOT initiate_chat or GroupChatManager
• Keep max_rounds reasonable (10-20)

When to Use This Pattern

• Unpredictable conversation flow where any specialist might be needed
• Brainstorming or collaborative problem-solving
• When defining explicit routing rules would be overly complex
• Natural team collaboration where agents join when their expertise is relevant

Example

See examples/organic_team.py for a complete project management team example.