How to build and scale multi-agent AI systems
summary
Shows how to deploy and secure agentic workloads on GKE with the Kubernetes-sigs Agent Sandbox (including a warm pool of sandboxes to avoid cold starts), use Gemini plus the Model Context Protocol (MCP) to get full infra visibility and automate diagnostics into reusable Agent Skills, and build a distributed multimedia pipeline on GKE with Gemini 3.5 Flash that extracts entities from Cloud Storage, writes a knowledge graph into BigQuery Property Graph, and queries it with GQL; demos include fixing an image-pull typo via an agent and judging 400+ hackathon projects using sandboxed code execution—end-to-end flows run successfully.
AI-written summary · about
chapters (6)
- 0:00Multi-agent AI systems intro
- 1:50Cloud engineers' AI toolkit
- 5:01Troubleshooting GKE with AI
- 32:29Secure untrusted code in GKE
- 59:34Knowledge graphs from pet data
- 1:30:13Q&A on AI agents
related talks
39mBuild the new generation of AI Agents
37mBuild agents with Gemini API
49mLet's build an AI agent
01h 48mBuilding Conversational Agents
14mTaming Agentic AI: Run Rogue Code Safely on Kubernetes with Agent Sandbox
30mSo You Want To Run AI Agents on Kubernetes: A 101 Guide
39mTrust, but Sandbox: Securing AI Agents in Kubernetes
25mWhat We Learned Securing AI Agents at Scale on Multi-Tenant Kubernetes Clusters