Merge CTO on Context Graphs: why hooking up MCP servers alone isn't enough for enterprise agents
AI Engineer · youtube · 2026-10-10
At AI Engineer World's Fair, Merge co-founder and CTO Gil Feig explained why enterprise agents need a "context graph" — a company brain combining third-party data, structured context, memories, and skills.
The problem: an agent can explain why one customer was upset, but asking which customers were upset last week breaks down — live MCP/API lookups can't search across large datasets.
The architecture:
- Sync a copy of data for semantic queries; keep live calls for simple lookups, with a router choosing between them
- A summarizer consolidates context
- Company skills and memory (e.g. how your company defines a "happy customer") belong in the context layer
- Four tiers of context, including derived summaries
- Traceability: every answer tracks where its data came from, when, and what invalidates it
Feig also shared a decision flowchart for cache vs. live data. "No magic, just physics."
More from coding & agent
- Claude cut a 2.6-hour video export to 20 minutes — but only after being told to rebuild the server — byhewar · 2026-10-23
- 3 flight-booking rules my AI agent checks that I'd never do by hand — Efistoffeles · 2026-10-23
- OpenJet claims Qwen3.8-27B coding runs 1.9x faster than llama.cpp on one RTX 3090 — JeremyCMorgan · 2026-10-23
- Building an autonomous SEO agent in Hermes with three skills: programmatic pages, Keywords Everywhere, DataForSEO — EXM7777 · 2026-10-23
- Building a sports betting AI agent with onchain execution and human approval by default: architecture and lessons — Dry-Comfortable-2678 · 2026-10-23
- How do you audit AI agent decisions in regulated workflows? A field survey — Ok_Way_1993 · 2026-10-23