Ben Horowitz: Databricks' agent system of record lifts accuracy from 50% to 84.5%
irinarish · x · 2026-10-28
Ben Horowitz's a16z essay argues the enterprise moat lies in what agents learn about a company's own business, not rented frontier models. Databricks' answer is a system of record answering four questions: what's true (Lakehouse), what it means (Genie Ontology — grounding lifted accuracy from 50% to 84.5% in their testing), what's allowed (Unity Gateway: identity, budget, guardrails), and whether it worked (Agent Bricks grading). Corrections feed back so learnings survive model swaps. Databricks has surpassed a $7B run rate, growing 80% y/y.
Related event: Databricks Tops $7B Revenue as Agent Queries Outnumber Humans, a16z Reports(3 posts)→
More from coding & agent
- CMU L3 Lab kicks off CUA Week with multiple realtime agent research releases — wellecks · 2026-10-28
- Team builds Jev-powered MCP router that cuts model costs up to 90% with 76% speedup — Fun-Meaning-6474 · 2026-10-28
- Odyssey Publishes Prompting Guide and Full Episode 1 Workflow, From Script to Edit — NicoVerderosa · 2026-10-28
- Nace AI's Drex for Search Tops Snowflake MADQA With 94.7% Accuracy — testingcatalog · 2026-10-28
- Drex for Search tops Snowflake MADQA leaderboard with 94.7% on 500 doc questions — testingcatalog · 2026-10-28
- Marc Schmidt: ultimate agent skills are flipped neurons, not markdown files — MarcJSchmidt · 2026-10-28