GraphQL Meets LLMs & Agents: Building Production AI at Starbucks Scale
summary
Starbucks describes using GraphQL at 180M+ queries/day across 10,000 stores and 31M+ app users as the substrate for AI agents, claiming 75x token cost reduction via selective field retrieval. Implementations include an in-store AI assistant on an Order Engine GraphQL BFF, a mobile/web platform on Apollo Supergraph, and on-call automation using Model Context Protocol (MCP) servers. The talk shows real demos, federation-based governance patterns, field-level auth, schema descriptions for tool discovery, and measured payload-efficiency and observability lessons for production.
AI-written summary · about
chapters (11)
- 0:00Introduction and speaker background
- 2:28Starbucks GraphQL journey
- 7:06AI changes the client model
- 9:11GraphQL’s original problems solved
- 11:20GraphQL accidentally built for AI
- 13:49API patterns for AI agents
- 15:58Payload efficiency in AI systems
- 18:52Production reality and governance
- 22:02Unlocking AI with GraphQL
- 24:11Where to start with AI and GraphQL
- 24:52Closing thoughts on GraphQL and AI
related talks
26mJust Ask: Building a Natural Language GraphQL Agent from the Graph You Already…
47mSchema Design in the Age of AI: Pragmatic Patterns to Evolve and Ship Fast
21mKeynote: How GraphQL is Redefining API Orchestration for the AI Era
01h 00mWorkshop: Getting Data Ready for AI - With Unified GraphQL &SQL Endpoints
26mAsk, Don't Fetch: Why GraphQL + MCP Is the Native Language of AI Agents
42mPanel - Fortune 100 Graphs - Powering Today's Need and Tomorrow's Trends
11mSponsored Keynote: GraphQL, BFFs and AI
24mScalable GenAI in Production