Snowflake Cortex Sense Enhances Enterprise AI Agents with Automatic Context
Snowflake is launching Cortex Sense, a new private preview feature designed to improve the accuracy and efficiency of AI agents by automatically building a contextual understanding of enterprise data. Unlike static semantic views, Cortex Sense dynamically models the entire business using signals like query history and BI metrics, addressing gaps in existing documentation and handling new data. This aims to reduce incorrect or speculative AI-generated answers for a broad range of users, from product managers to data scientists, across various cloud platforms.
- →Cortex Sense provides automatic context for AI agents
- →Automated self-correction and conflict resolution
- →Role-based access control for context
- →Improved accuracy and efficiency for AI agents
- →Faster setup compared to manual processes
Features (3) ›
- Cortex Sense provides automatic context for AI agents
Cortex Sense is a new managed solution that builds an understanding of enterprise data to improve AI agent accuracy, working alongside existing semantic views. It automatically infers context from signals like past analyst queries, transformation tool models, and BI metrics, rather than relying solely on manually curated documentation.
- Automated self-correction and conflict resolution
The system includes a self-correcting loop to identify and resolve knowledge gaps and conflicts, involving human input when ambiguities arise. It ranks conflicting information based on relevance, authority, popularity, and freshness, similar to web search ranking.
- Role-based access control for context
Access to Cortex Sense context will be governed by roles within Snowflake, starting with a single role in private preview and planning to expand to per-role contexts for different teams like marketing and finance.
Enhancements (1) ›
- Improved accuracy and efficiency for AI agents
Cortex Sense aims to significantly increase AI agent accuracy, with internal benchmarks showing an improvement from 24.1% to 86.3% on complex queries. It also reduces per-query costs by preventing agents from inefficiently inspecting every table.
Notes (1) ›
- Faster setup compared to manual processes
Setting up Cortex Sense is designed to be significantly faster than traditional manual documentation or semantic view creation projects, with internal setups taking a single day rather than months.
https://www.snowflake.com/content/snowflake-site/global/en/blog/enterprise-ai-agents-grounded-context