Snowflake Cortex Sense Auto-Builds Semantic Layers to Fix Text-to-SQL Accuracy

Snowflake has introduced Cortex Sense, a tool designed to address the longstanding unreliability of natural language-to-SQL conversion in enterprise AI. The core issue is not syntax but context — AI models lack the implicit business knowledge needed to correctly interpret ambiguous queries, with baseline text-to-SQL accuracy sitting at just 21–25% without a context layer. Rather than requiring teams to manually curate a semantic layer, Cortex Sense automatically builds one by observing existing analyst queries, transformation models, and BI dashboard metrics via Snowflake Horizon Connectors. The automated approach significantly reduces setup bottlenecks and allows the semantic model to evolve as business data usage changes. However, the system carries a trade-off: if past queries or dashboards contain errors or outdated logic, Cortex Sense risks learning and propagating those inaccuracies at scale without human oversight.