Snowflake Unveils Business Ontology to Give AI Agents Approved Definitions

One official place for business meaning

According to Snowflake, Business Ontology is part of Snowflake Horizon Context, the governed context layer in Horizon Catalog. Teams can reuse the same knowledge across agents and applications without redefining it for each tool, while keeping differences across domains such as sales, finance and marketing.

Snowflake describes the problem with an example: an agent answering a revenue question needs to know how finance defines "revenue" and how it relates to orders and customers. That knowledge, the company says, usually lives across BI models, dbt projects and the experience of long-tenured employees, and teams often rebuild it for each new agent. If an agent gets it wrong, people make decisions based on the wrong number and lose trust in AI answers.

In Business Ontology, an analyst can define a term such as "net revenue" once, or import it from an existing glossary. The analyst sets its scope to "finance," names an owner, gets it approved and links it to the semantic view metric that calculates it. From then on, agents and other teams work from that approved definition.

Three layers of context

Snowflake offers three layers of context, each with a distinct role:

• Snowflake Cortex Sense (public preview soon) infers context automatically.
• Business Ontology governs business meaning.
• Snowflake Semantic Views execute metrics.

The layers are not sequential. Customers can start with whichever one addresses their most pressing problem and add the others as they grow. Some teams may begin with glossary needs, others with executable metrics.

Cortex Sense builds context from metadata and usage patterns teams already produce, including query history, dashboards and dbt models, and updates its knowledge at a chosen cadence. It flags conflicting knowledge and asks teams which source to trust. When teams are ready to govern key terms, it can create an initial draft in Business Ontology with proposed definitions, relationships and links to data. When an approved definition conflicts with something Cortex Sense inferred, agents use the approved definition and Cortex Sense learns from it, according to Snowflake.

Semantic Views package a domain's metrics, dimensions and joins into a Snowflake object that agents, BI tools and SQL can query directly. Business Ontology links approved definitions to these views but does not change the calculation, which stays with each view's owners.

Governance, variants and connectors

Snowflake says teams can ground agents in business concepts, align metrics across semantic views and help users find the right definition. Anyone can look up a term, see who owns it and whether it is approved, and follow it to where it is calculated.

Intentional differences are preserved. Snowflake gives the example of a restaurant company that defines same-store sales differently for each brand, with each version approved separately, labeled with its brand and linked to its own semantic view.

Organizations control who can create, edit and approve definitions through Horizon Catalog's governance controls. AI suggestions and imported definitions are a starting point for review, and accountable owners decide what becomes official.

Horizon Context also draws on existing tools. Its metadata connectors collect schemas, query logs, dashboard definitions and usage signals from sources such as Tableau, Power BI, dbt, SQL Server, Sigma, Omni and PostgreSQL. Cross-platform lineage traces data from a source database through transformations to the dashboard that displays it.

Open standards and early feedback

Snowflake says it helped found Apache Ossie (incubating), a vendor-neutral open standard for semantic models backed by dozens of organizations, including Microsoft, Salesforce, dbt Labs, Qlik and ThoughtSpot. Semantic Views can import and export Ossie definitions, so metrics can move between Snowflake and other compatible tools. Agents outside Snowflake can also query Semantic Views through Model Context Protocol, with Horizon Catalog governing access.

Joe Tobey, Head of Data Product Engineering at dentsu, said as an early product adopter: "this is what lets us put agents in front of our data with confidence." Morten Lileng, Global Head of Engineering at dentsu, said Business Ontology gives "every definition an owner and every downstream system a single governed answer."

What it means for users

For organizations deploying AI agents on enterprise data, the pitch is consistency: one approved, owned definition that agents and people share, instead of separate versions rebuilt for each tool. Snowflake has not yet given a public preview date beyond "soon" for Business Ontology or Cortex Sense.