Building Autonomous Agents with Python & AutoGen

Single LLMs are smart. Teams of LLMs are geniuses. Microsoft's AutoGen allows you to spawn multiple agents that talk to each other.

The Concept

Imagine a virtual software company:

1. Product Manager Agent: Defines the requirements.
2. Coder Agent: Writes the code.
3. Reviewer Agent: Checks the code for bugs.

Installation

pip install pyautogen

The Code

from autogen import AssistantAgent, UserProxyAgent, config_list_from_json

# Configuration
config_list = config_list_from_json(env_or_file="OAI_CONFIG_LIST")

# Create Assistant (The Coder)
assistant = AssistantAgent(
    name="coder",
    llm_config={"config_list": config_list}
)

# Create User Proxy (The Executor)
user_proxy = UserProxyAgent(
    name="user_proxy",
    code_execution_config={"work_dir": "coding", "use_docker": False}
)

# Start the chat
user_proxy.initiate_chat(
    assistant,
    message="Plot a chart of NVDA stock price year-to-date. Save it as nvda.png."
)

What Happens?

1. Coder writes a Python script to fetch data and plot.
2. User Proxy executes the script on your machine.
3. If it fails, User Proxy sends the error back.
4. Coder fixes the bug and sends new code.
5. Loop continues until success.

Conclusion

AutoGen represents the future of software development: Humans supervising a digital workforce.