How to build and scale multi-agent AI systems on GKE

Learn how to deploy and secure agentic workloads on Google Kubernetes Engine (GKE) using the Kubernetes-sigs Agent Sandbox for automated application evaluation. Discover how to troubleshoot broken GKE deployments using Gemini and the Model Context Protocol (MCP) to gain full infrastructure awareness and automate diagnostic workflows into reusable skills.

Additionally, explore how to build a distributed multimedia knowledge acquisition pipeline on GKE using Gemini 3.5 Flash to extract entity information from Cloud Storage assets. Store the resulting knowledge graph data in BigQuery and analyze relationships using BigQuery Property Graph and Graph Query Language (GQL).

*What you'll learn:*
* How to initialize GKE Autopilot and install Kubernetes-sigs Agent Sandbox CRDs and components.
* How to troubleshoot GKE deployments using Gemini combined with the GKE Model Context Protocol (MCP) server.
* How to codify infrastructure diagnostic workflows into reusable Agent Skills.
* How to create and query a Property Graph in BigQuery using Graph Query Language (GQL).

🔗 *Resources:*
* Cloud Engineering's AI Toolkit: Platform Engineering on GKE using Gemini → https://g.dev/ai/ai-toolkit-gke-1
* Automatic Code Evaluation with Agent Sandbox on GKE → https://g.dev/ai/ai-toolkit-gke-2
* Scale Distributed Data Processing with GKE to build a Knowledge Graph in BigQuery → https://g.dev/ai/ai-toolkit-gke-3

Speakers: Mofi Rahman, Lucia Subatin, Olivier Bourgeois
Products Mentioned: Google Kubernetes Engine, Gemini, BigQuery