Snowflake expands AI agent capabilities and governance

At its Expedition 2026 event, Snowflake announced new features designed to expand the functionality of AI agents while increasing enterprise controls. The updates focus on three core areas: grounding agents in authoritative data, enabling them to perform actions within user permissions, and compounding organizational knowledge.

To address issues regarding data accuracy and consistency, Snowflake introduced the Business Ontology. This tool provides agents with approved definitions for business concepts, such as net revenue, and connects them to underlying data tables. This is intended to prevent agents from returning incorrect figures due to a lack of context.

The Cortex AI Gateway is now generally available, offering governance over agent connections, activity records, and spending controls to enforce budget limits. Other new features include MCP Tools, which is in public preview, and several tools currently in private preview, such as Agent Abuse Detection and Agent Trace Intelligence.

Snowflake is also deploying CoCo, a coding agent for data teams, to assist with recurring pipeline maintenance. While some capabilities are generally available, other features, including proposed always-on bots, remain in preview stages.