Snowflake Announces Business Ontology for Governed AI-Agent Context

Snowflake adds governed business context for AI agents

Snowflake announced Business Ontology on November 3, describing it as a governed way for teams to define, own and approve shared business terms, entities, relationships and metrics and connect those definitions to the data and models implementing them. The feature is part of Snowflake Horizon Context in Horizon Catalog and is scheduled for public preview soon; the reviewed announcement did not establish general availability.

Business Ontology is intended to give people and AI agents the same approved definitions for enterprise data. Snowflake says teams can reuse the resulting context across agents and applications while retaining differences among domains such as sales, finance and marketing. The company uses net revenue as an example of a definition that may need to connect a finance-approved meaning with related orders and customers.

One catalog for approved definitions and data

The announced capability lets organizations assign ownership to business concepts, connect them to underlying tables and analytical views, and record changes for audit purposes. Its objective is to replace scattered definitions in BI models, dbt projects and data dictionaries with a central record that can be reviewed and approved by business users.

Snowflake said Business Ontology builds on Horizon Context, whose metadata connectors gather information such as schemas, query logs, dashboard definitions and usage signals from systems including Tableau, Power BI, dbt, SQL Server, Sigma, Omni and PostgreSQL. Cross-platform lineage is intended to trace data from source databases through transformations to dashboards.

The company is positioning Business Ontology alongside two other context capabilities. Cortex Sense, also slated for public preview soon, is intended to infer context automatically, while Semantic Views execute metrics in a consistent manner. Snowflake said the three layers are not sequential and that customers can begin with whichever layer addresses their most immediate requirement.

Reliability depends on ownership and upkeep

The announced approach adds a human review stage to meanings that might otherwise be inferred from metadata and usage patterns. Researchers cited in the reviewed reporting said central definitions could reduce the need to embed the same business logic in every agentic application and limit conflicting interpretations across tools.

That benefit does not remove the work of keeping definitions current. A definition for net revenue, active customers or another core term must be reviewed when products, pricing, accounting rules or reporting requirements change. Schema changes may also require someone to check affected definitions. Differences among departments—such as sales and finance using different customer definitions—will still need to be reconciled or maintained as separate, clearly identified terms.

Snowflake cited research it sponsored with MIT Technology Review Insights in which 55% of executives identified a lack of business context as the main cause of poor-quality AI-agent insights, while 45% pointed to fragmented or inaccessible data. The survey establishes the views of the executives surveyed and does not independently validate the performance of the announced product.

Governance controls are separate from context

Snowflake also introduced controls for the Cortex AI Gateway. Reported features include isolated environments for agent-written code, tools for inspecting execution traces and mechanisms intended to identify agents acting outside permitted parameters. Business Ontology itself was described as a public preview coming soon, while the reported gateway-control availability differed by feature.

The separation matters because a governed definition can establish what a business term means without establishing that an agent’s code or output is correct. Snowflake’s broader thesis is that agents require approved context, controlled data access and accountability for their actions as enterprise deployments expand. The reviewed material did not provide evidence of completed customer procurement or quantified results from the new capabilities.