https://video.ut0pia.org/videos/watch/69080589-bd12-43e2-b36d-68223baf35c9 Models, frontends, and agent frameworks all commoditize. Mike Phipps argues the durable moat is the one thing that does not: your data model, and the tacit knowledge of how your questions are supposed to be answered. At the Gates Foundation he and his team modeled 25 years of grantmaking, over 7 billion dollars a year across 2,000 grants and 4,000 people, into a single Neo4j knowledge graph served to Claude through one MCP server. The graph is built for agents, not dashboards: hierarchies become traversable paths, and unstructured documents are chunked, tagged, and mapped to structured entities at ingestion. An agent turns a messy cross system question into one graph query and gets an answer that respects how the organization actually reports. The talk walks the architecture and the retrieval evals that keep it honest, and why a small team's effort compounds in the data model, not the layers above it. Speaker info: https://www.linkedin.com/in/mike-phipps-79339a38, Timestamps: 0:00 - The moat question: what stays defensible as AI commoditizes 2:32 - SIP across the 4,000 person Gates Foundation 3:46 - The scale: 25 years, $7B a year, 2,000 grants 5:15 - Structuring operational data for agentic retrieval 6:18 - Tacit knowledge as the moat: engaging data owners 7:09 - The curation pipeline and governance 8:30 - Modeling hierarchies: funding and management lenses 12:32 - People, org charts, and stitching siloed systems 13:49 - Combining unstructured documents with structured data 15:34 - Serving the graph to Claude through MCP 17:31 - Retrieval evals and the feedback loop Thu, 23 Jul 2026 09:52:15 GMT https://validator.w3.org/feed/docs/rss2.html PeerTube - https://video.ut0pia.org https://video.ut0pia.org/lazy-static/avatars/0287a09a-aae7-4840-9843-b416426e7046.webp https://video.ut0pia.org/videos/watch/69080589-bd12-43e2-b36d-68223baf35c9 All rights reserved, unless otherwise specified in the terms specified at https://video.ut0pia.org/about and potential licenses granted by each content's rightholder.