Snowflake Adds Business Ontology and Expands AI Agent Governance at Expedition 2026

The main additions are a Business Ontology framework, wider agent controls in Cortex AI Gateway, and data engineering tools that let AI agents build, monitor and fix data pipelines with less human input.

Shared Definitions for Agents

Business Ontology builds on Snowflake's existing semantic views. It lets a company set approved definitions of business concepts and apply them across the organisation. Snowflake's argument is that without a shared definition, two agents can give conflicting answers to the same business question even when both are working from accurate data. Snowflake's coding agent for data teams, CoCo, and its personal work agent for knowledge workers, CoWork, will use those approved definitions to ground their answers. The feature is in private preview and is due to move to public preview soon.

Chief executive Sridhar Ramaswamy said AI is "industrializing intelligence" and that work which "once took months can now take hours, days, or even minutes." He argued that as business users become developers, competitive advantage moves from specialist skills to a company's own knowledge.

Governing Agents Like Employees

Cortex AI Gateway launched in July 2026 as a central point for governing and monitoring agents. The new version adds agent abuse detection, observability and tracing, Cortex sandboxes, AI cost governance, dynamic model routing, and tool integration that uses technology from Snowflake's acquisition of Natoma. It applies existing security controls, such as data masking and access restrictions, to actions taken by agents and keeps records of what those agents do. IT and security teams can set cost limits, choose models for efficiency or accuracy, and control in detail which external tools agents may use.

"An agent that acts on your business has to be trusted, and it has to know three important things: what the data means, which data to trust, and what it's allowed to do," Ramaswamy said.

Snowflake is competing to own the control layer for enterprise agents. SAP has been expanding its Autonomous Enterprise architecture and its governed Joule agents. Salesforce has introduced an Enterprise AI Harness and an AI Control Plane. Snowflake is approaching the problem from the data platform side.

Smaller Models and an Oracle Tie-Up

Snowflake also put Cortex Training into public preview. The service lets customers post-train open-weight models on their own data without managing GPU infrastructure. Executive vice president of product Christian Kleinerman said specialised, smaller models can give faster responses and often lower token costs. SiliconANGLE noted that the service may appeal to companies that want more control over their models because of data-handling rules or restrictions on foreign open-weight AI.

Separately, Snowflake expanded its partnership with Oracle, which will make the Snowflake platform available as a first-party service on Oracle Cloud Infrastructure. Customers will be able to use Snowflake's analytics and AI features directly through Oracle databases and applications.

For chief data and technology officers, the announcements point to the next stage of enterprise AI spending. Companies are moving past model access and are now paying for controls that make agent behaviour consistent, auditable and affordable.