Big Graphs, Tiny Contexts: Dev Tools for Agents
summary
Netflix runs a federated enterprise graph about 4 Shakespeares (~4M tokens) spanning roughly 1,000 subgraphs from hundreds of teams, so agents can never ingest the whole schema. The platform exposes a remote MCP plus a GraphQL schema registry, a hand‑built schema search index (Elasticsearch) to return narrow type slices, and a deterministic query validation API that gates returned queries. They favor server‑side slicing, CLIs for local files, guardrails to prevent hallucinations, and have shelved naive RAG for now.
AI-written summary · about
chapters (10)
- 0:00Introduction and setup
- 1:27Netflix’s two federated graphs
- 5:37Developer onboarding challenges
- 8:03Exposing tools to agents
- 11:56GraphQL MCP architecture
- 14:29Schema search with metadata
- 15:54Query validation API
- 16:59Agents inside development tools
- 20:33APIs as the foundation
- 22:00Key takeaways and Q&A
related talks
40mAgent-Ready Graphs: Engineering the Foundations for Discoverability, Safety, and QoS
07mshare how avoid these GraphQL pitfalls
15mMultiple Knowledge Graphs for Sophisticated RAG
26mJust Ask: Building a Natural Language GraphQL Agent from the Graph You Already…
14mUsing Graphs to Fine-Tune Agents
53mAgents, Tools, and Mcp, Oh My! Next Level AI Concepts for Developers
29mClosing the Loop: How GraphQL Gives Coding Agents Eyes on What Actually Matters
30mBuilding Smarter Mobile Apps with On-device AI Agents & GraphRAG