Snowflake adds business context and controls for enterprise agents

Snowflake expanded its AI platform with capabilities designed to help enterprises deploy agents grounded in proprietary knowledge and governed by existing policies. The announcements include Business Ontology, expanded Cortex AI Gateway controls and autonomous data engineering tools that let agents build, monitor and troubleshoot pipelines with less human intervention.

Business Ontology extends Snowflake’s semantic views with approved organizational definitions, helping prevent agents from producing incompatible answers from accurate underlying data. Snowflake CoCo, its coding agent for data teams, and Snowflake CoWork, its personal work agent for knowledge workers, use those concepts to ground responses. The framework is in private preview, with public preview planned.

Cortex AI Gateway gains abuse detection, observability and tracing, sandboxes, cost governance, dynamic model routing and tool integration using technology from Snowflake’s Natoma acquisition. Existing data masking and access restrictions apply to agent operations, while security teams can monitor activity, maintain records and restrict external tools.

Snowflake also announced the public preview of Cortex Training, which lets customers tailor open-weight models using proprietary information without managing the underlying infrastructure. The service targets specialized models that can deliver faster, more efficient responses. An expanded Oracle partnership will make Snowflake available as a first-party service on Oracle Cloud Infrastructure, enabling access to its analytics and AI capabilities through Oracle databases and applications.