Multimodal, Real-Time AI Agent Systems

Chapter 6. Multi-Agent Systems: Collaboration and Orchestration (Illustrated with ADK)

A Note for Early Release Readers

With Early Release ebooks, you get books in their earliest form—the author’s raw and unedited content as they write—so you can take advantage of these technologies long before the official release of these titles.

This will be the 7th chapter of the final book. Please note that the GitHub repo will be made active later on.

If you’d like to be actively involved in reviewing and commenting on this draft, please reach out to the editor at [e-mail].

In the last chapter, you got your hands dirty and built a complete, capable agent from the ground up. You saw how to give it tools, make it “live” with real-time audio , and even how to test it using ADK Web. That single-agent system is powerful, but as you scale, you’ll run into the same challenges that software developers face with large, monolithic applications.

A single, “do-everything” agent can become a bottleneck. As you pile on more and more tools for different, unrelated tasks (like math, history, and grammar), the agent’s core reasoning can get confused, its performance can drop , and it can become a nightmare to test and maintain. As you can see in Figure 6-1, while single agents are simpler to start with, they can get stuck in loops and perform poorly when their responsibilities grow too large.

Figure 6-1. A comparison of single-agent and multi-agent architectures, highlighting their respective ...