Multi-agent systems, concepts & patterns | The Agent Factory Podcast

Published on 14 Jul 2025, 23:00

Ready to move beyond single-agent limitations? This episode of The Agent Factory is your deep dive into designing and building powerful multi-agent systems. Join hosts Vlad Kolesnikov and Shir Meir Lador as they explore the architectural patterns that enable agents to collaborate on complex tasks, from hierarchical supervisors to dynamic swarms. By the end of this episode, you'll have a solid understanding of the critical concepts of Context Engineering and how to choose the right multi-agent pattern for your project. We'll break down the practical trade-offs of each architecture so you can design more capable and resilient agentic systems.

In This Episode, You'll Learn:

Agent Industry Pulse: We explore the paradigm shift to "Software 3.0" and why the LLM is becoming the new OS. We'll cover the rise of "Context Engineering" as a critical discipline for managing an agent's "RAM" and discuss new open-source tools like the Gemini CLI for developers and the A2A (Agent-to-Agent) protocol for standardized agent communication.

The Factory Floor: We answer the core question: when should you use a multi-agent system? We'll provide a detailed breakdown of key architectural patterns, including:

- Supervisor Patterns: (Router & Parallel) for managed, hierarchical task delegation.
- Deterministic Flows: (Sequential & Circular) for predictable, assembly-line-style processing.
- Dynamic Swarms: for flexible, all-to-all agent collaboration.
- Live Demo: Vlad demonstrates a multi-agent system in action, showcasing the Router/Supervisor pattern.

Developer Q&A: We tackle a critical community question on gaining observability into multi-agent systems. Learn how to debug agent delegations and decisions using tools like Google Cloud Trace Explorer and libraries like OpenTelemetry to make your agent's reasoning transparent.

About The Agent Factory:
"The Agent Factory" is a video-first technical podcast for developers, by developers, focused on building production-ready AI agents. We'll explore how to design, build, deploy, and manage agents that bring real value using the latest open-source tools and frameworks. Going beyond basic AI/ML concepts, the podcast will delve into the practical aspects of agentic AI, covering everything from single-purpose agents for task automation to complex multi-agent systems.

Connect with Us & Ask Questions:
- Leave your questions and feedback in the YouTube comments below!
- Join the conversation on social media with the hashtag #TheAgentFactory.
- Connect with the community at the Google Cloud Community forum → goo.gle/3IxBV2D

Resources:
Software 3.0 & Vibe Coding:
ZDNet Article → goo.gle/3IrzdM9
Andrej Karpathy on LLM OS → goo.gle/4nSPsli
Vibe Data Analytics Agent → goo.gle/cloud-crm-data-agent

Context Engineering:
Letta Blog - Memory Blocks → goo.gle/40m9kTU
ArXiv Paper on Context Management → goo.gle/40XPkHk
Langchain blog on context engineering → goo.gle/40m9ElA
Rlance Martin Blog → goo.gle/40pHfep
LangGraph Context Isolation Demo → goo.gle/3INhdM1

Tools & Protocols:
Gemini CLI→ goo.gle/3UerbZk
Gemini CLI demos and tutorials → goo.gle/4lQTI2U
A2A Protocol Donation to Linux Foundation → goo.gle/4lzK9G4
A2A Multi-Language Sample (Python/Java) → goo.gle/469OE5i
A2A Inspector Tool → goo.gle/4lWSLGv

Observability & Tracing:
Google Cloud Trace Docs → goo.gle/4eZCq1G

Subscribe to Google Cloud Tech → goo.gle/GoogleCloudTech

#MultiAgentSystem #ContextEngineering #GoogleCloud

Speakers: Vlad Kolesnikov, Shir Meir Lador
Products Mentioned: Cloud General, Gemini, Agent Developer Kit (ADK), Cloud Run