Agent-Ready Graphs: Engineering the Foundations for Discoverability, Safety, and QoS
summary
Booking.com builds an agent-ready federated supergraph with a schema QA pipeline backed by expert review and entity modeling, enforced owner consent for MCP-registered operations, and field-level policy directives for access control. Platform runs >170 subgraphs, ~8,000 tabs, 4,000 live operations and ~80% domain coverage. They ship a text-to-GraphQL agent (developer-facing, ~1 minute per complex query) and an agent flow that demos booking in staging. Tooling generates ~500 select-star queries then chunks to ~60–70 pieces, stores embeddings in Bedrock + OpenSearch, detects PII and enforces consent scopes in access tokens.
AI-written summary · about
chapters (12)
- 0:00Agent-ready graphs at Booking.com
- 1:09Booking.com’s GraphQL evolution
- 4:39Supergraph scale and complexity
- 6:03Text to GraphQL demo
- 7:51Agentic trip booking demo
- 9:41Journey to agent-ready graphs
- 14:20Schema QA for AI agents
- 18:34Tooling for GraphQL schemas
- 26:34Agent access security
- 30:08Hyper-personalized UI plans
- 31:10Product capabilities for AI agents
- 31:54Q&A
related talks
23mBuilding the Graph at Booking.com
23mBuilding the Graph at Booking.com
29mBig Graphs, Tiny Contexts: Dev Tools for Agents
26mJust Ask: Building a Natural Language GraphQL Agent from the Graph You Already…
29mScaling the Data Access by Leveraging GraphQL Federation for Service to Service Communication
27mHow GraphQL Gives Coding Agents Eyes on What Actually Matters
30mDeep Learnings Migrating legacy services to Federated GraphQL
35mGraphQL in the Enterprise