The useful question is what changes for users, developers or buyers, and whether the announcement stays industry context or becomes something people can actually use.

Snowflake introduced Cortex Sense to address the limitations of traditional enterprise data documentation, which often lags behind business changes and covers only a small fraction of available tables. Unlike semantic views, which require manual curation and cover less than 5% of Snowflake’s 9,685 internal tables, Cortex Sense automatically builds an understanding of the entire data estate by analyzing existing signals such as query history, transformation models, and BI metrics. This approach aims to reduce reliance on outdated or incomplete documentation while maintaining governance where needed.

Cortex Sense operates alongside semantic views, treating them as authoritative sources for governed answers while independently modeling the broader data context. It mitigates risks of confidently incorrect responses by running a self-correcting loop that identifies gaps and conflicts, surfacing ambiguities to users for resolution. The system ranks conflicting signals by relevance, authority, and freshness, prioritizing governed definitions over inferred ones. Access is controlled through Snowflake’s existing role-based governance, with initial private preview allowing a single role to access all context, expanding to per-role contexts later.

Internal testing showed Cortex Sense improved AI agent accuracy from 24.1% to 86.3% on benchmark questions, while reducing query costs from $1.76 to $0.59 per query by eliminating unnecessary table inspections. The system achieved parity with hand-curated semantic views on early evaluations and surpassed them by 10 percentage points in areas with stale definitions or uncovered domains, thanks to signals like recent query history. Setup requires minimal effort, with indexing completed in a single day and no manual curation needed.

Cortex Sense will enter private preview in mid-July 2025, with early access available by contacting Snowflake account representatives. The product is positioned as a shift from periodic expert-driven documentation to continuous, self-updating context learning, complementing semantic views rather than replacing them. Snowflake emphasizes its managed nature, reducing maintenance overhead while improving AI agent reliability across evolving enterprise data estates.