Cortex Sense for Enterprise AI Agents | Snowflake

Details

Snowflake's Cortex Sense addresses a practical data governance problem: enterprise AI agents need semantic understanding of company data to give accurate answers, but manual semantic view creation has not scaled. Snowflake's own product team had documented fewer than 5% of its 9,685 tables with semantic views before the bottleneck became acute — most questions from users fell outside the documented slice. Cortex Sense is designed to infer semantic context automatically from signals the business already produces: analyst query history, transformation tool models (such as dbt definitions), and BI metrics. It builds a working model of the entire data estate rather than just cataloging tables. Integration comes via Snowflake Horizon Connectors. Existing semantic views remain the gold standard for governed, consistent answers — Cortex Sense supplements rather than replaces them. The feature directly benefits teams running Cortex Agents, Snowflake CoCo, or third-party agents like Cursor or ChatGPT that query Snowflake tables — which currently may produce confident but incorrect answers over undocumented data. Private preview begins mid-July 2026 with no GA date announced.