4.4. Autogen Multi-Agent Systems

💡 First Principle: Some tasks benefit from debate and specialization—a researcher finds facts, a writer drafts content, an editor refines it. Autogen models this as agents in a group chat, each with a persona and expertise, talking until they converge on a solution. The GroupChatManager acts as a moderator deciding who speaks next. Use Autogen when your task naturally decomposes into roles that benefit from back-and-forth refinement, not just sequential steps.

Core Agent Types

🔧 Implementation Reference: Autogen

Two-Agent Conversation

The simplest Autogen pattern pairs an AssistantAgent with a UserProxyAgent.

Testable Pattern:

from autogen import AssistantAgent, UserProxyAgent

# Create AI assistant
assistant = AssistantAgent(
    name="assistant",
    llm_config={
        "config_list": [{
            "model": "gpt-4o",
            "api_type": "azure",
            "api_key": "your-key",
            "base_url": "https://your-resource.openai.azure.com/",
            "api_version": "2024-08-01-preview"
        }]
    },
    system_message="You are a helpful coding assistant."
)

# Create user proxy (can execute code)
user_proxy = UserProxyAgent(
    name="user_proxy",
    human_input_mode="NEVER",  # Fully autonomous
    code_execution_config={
        "work_dir": "coding",
        "use_docker": False
    }
)

# Start conversation
user_proxy.initiate_chat(
    assistant,
    message="Write a Python function to calculate the Fibonacci sequence."
)

Human Input Modes

Multi-Agent Group Chat

For complex tasks, multiple specialized agents collaborate in a group chat.

Group Chat Pattern:

from autogen import AssistantAgent, GroupChat, GroupChatManager

# Create specialized agents
planner = AssistantAgent(
    name="Planner",
    system_message="You break down complex tasks into steps.",
    llm_config=llm_config
)

coder = AssistantAgent(
    name="Coder",
    system_message="You write Python code to implement solutions.",
    llm_config=llm_config
)

reviewer = AssistantAgent(
    name="Reviewer",
    system_message="You review code for bugs and improvements.",
    llm_config=llm_config
)

# Create group chat
groupchat = GroupChat(
    agents=[planner, coder, reviewer],
    messages=[],
    max_round=12,
    speaker_selection_method="auto"  # LLM decides who speaks next
)

# Create manager to orchestrate
manager = GroupChatManager(
    groupchat=groupchat,
    llm_config=llm_config
)

# Start multi-agent conversation
planner.initiate_chat(
    manager,
    message="Build a REST API for a todo list application."
)

Speaker Selection Methods

⚠️ Exam Trap: human_input_mode="NEVER" enables fully autonomous agent execution without human intervention.

Autogen Documentation