How to Build Autonomous AI Agents That Run Anywhere
summary
Builds a stateful travel-planning agent using the Cloudflare Agent SDK, the Model Context Protocol (MCP) to call external tools (Amandus for flights, booking APIs), and Durable Objects (SQLite-backed) for persistent, immediately consistent memory. Authentication uses OIDC providers like Auth0/Stytch/WorkOS to mint bound MCP tokens per user. Demo shows scheduled reminders survive restarts, websocket hibernation sleeps idle MCP instances, and a mocked flight booking completes after a human approval step.
AI-written summary · about
chapters (17)
- 0:00Introduction and speaker intro
- 3:46Rise of AI agents explained
- 6:36Hard parts of building agents
- 7:19Introducing Cloudflare Agents
- 8:02Agent SDK features overview
- 9:09Four parts of an AI agent
- 12:04Workflows and tools in agents
- 13:54Building MCP servers on Cloudflare
- 14:37Solving identity challenges
- 17:30Memory and state management
- 19:57Agent coordination and hibernation
- 20:40Live demo of travel agent
- 24:54Code walkthrough and tools
- 26:47Scheduling and reminders
- 28:09Client and UI integration
- 29:55Agent architecture summary
- 30:16Getting started with agent SDK
related talks
11mCreate AI Agents That Work Independently
01h 30mHow to build AI Agents: A step by step guide to creating AI agents
30mComposable AI: Building Next-Gen AI Agents with MCP
58mAdvanced Autonomous AI Agents: Memory, Collaboration and Evolution in Action
36mPutting AI Agents to Work: A New Era of Open Source Connectivity with MCP
40mIn AI We Trust? Securing the Future, One Agent at a Time
19mThe Future of AI
34mAI Agents: Transforming Autonomous Decision-Making