How to Intermediate · 3 min read

How to orchestrate multiple AI agents

Quick answer

To orchestrate multiple AI agents, use a central controller program that manages communication and task delegation between agents via API calls or message passing. Implement each agent as an independent LLM instance (e.g., gpt-4o, claude-3-5-sonnet-20241022) and coordinate their inputs and outputs to achieve complex workflows.

PREREQUISITES

Python 3.8+
OpenAI API key (free tier works)
pip install openai>=1.0

Setup

Install the OpenAI Python SDK and set your API key as an environment variable to securely authenticate requests.

bash

pip install openai

Step by step

This example demonstrates orchestrating two AI agents: one generates a question, and the other answers it. The controller coordinates their interaction.

python

import os
from openai import OpenAI

client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])

# Agent 1: Generate a question
question_response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Generate an interesting trivia question."}]
)
question = question_response.choices[0].message.content
print(f"Agent 1 (Question): {question}")

# Agent 2: Answer the question
answer_response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": f"Answer this question: {question}"}]
)
answer = answer_response.choices[0].message.content
print(f"Agent 2 (Answer): {answer}")

output

Agent 1 (Question): What is the tallest mountain in the solar system?
Agent 2 (Answer): The tallest mountain in the solar system is Olympus Mons on Mars.

Common variations

You can extend orchestration by adding more agents, using asynchronous calls for parallelism, or integrating different models like claude-3-5-sonnet-20241022 for specialized tasks. Streaming responses can improve responsiveness in interactive setups.

python

import asyncio
import os
from openai import OpenAI

client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])

async def agent_generate():
    response = await client.chat.completions.acreate(
        model="gpt-4o",
        messages=[{"role": "user", "content": "Generate a creative story prompt."}]
    )
    return response.choices[0].message.content

async def agent_expand(prompt):
    response = await client.chat.completions.acreate(
        model="claude-3-5-sonnet-20241022",
        messages=[{"role": "user", "content": f"Expand this prompt into a short story: {prompt}"}]
    )
    return response.choices[0].message.content

async def main():
    prompt = await agent_generate()
    print(f"Agent 1 prompt: {prompt}")
    story = await agent_expand(prompt)
    print(f"Agent 2 story: {story}")

asyncio.run(main())

output

Agent 1 prompt: A mysterious door appears in the middle of a bustling city.
Agent 2 story: In the heart of the city, a door materialized overnight, shimmering with an otherworldly glow... (story continues)

Troubleshooting

If you see rate limit errors, implement exponential backoff retries or reduce request frequency.
Ensure environment variables are correctly set to avoid authentication failures.
Check model availability and update model names as APIs evolve.

✅

Key Takeaways

Use a central controller to manage communication between multiple AI agents via API calls.
Leverage different models for specialized tasks to build collaborative AI workflows.
Implement async calls to run agents in parallel and improve efficiency.
Handle API rate limits and authentication errors proactively.
Keep model names and SDK usage up to date with provider documentation.

Verified 2026-04 · gpt-4o, claude-3-5-sonnet-20241022

Verify ↗