Snowflake Debuts Business Ontology to Improve AI Agent Accuracy

Snowflake unveiled Business Ontology, a data governance tool that gives AI agents access to approved business definitions to reduce inconsistent answers, expected to enter public preview soon as part of Snowflake Horizon Context within the Horizon Catalog platform. Business Ontology works alongside Cortex Sense, which infers relationships from metadata and query histories, and Semantic Views, which handles metric calculations, and supports open formats including Apache Ossie and Resource Description Framework. Snowflake also expanded Cortex AI Gateway with Cortex Sandboxes and Agent Trace Intelligence, added Agent Abuse Detection in private preview, and made CoCo Cloud Agents and CoCo Automations generally available.

TL;DR — Key Takeaways - Snowflake’s Business Ontology gives AI agents access to approved business definitions to help reduce inconsistent answers. - Business Ontology works alongside Cortex Sense and Semantic Views to connect inferred context, approved meaning and consistent calculations. - Snowflake is expanding agent security, monitoring and cost controls, alongside CoCo tools for cloud-based coding and scheduled workflows. Snowflake has unveiled Business Ontology, a new data governance tool geared to improve the accuracy of AI agents by giving them access to approved business definitions. The offering is part of a larger Snowflake initiative to help companies deploy AI agents more reliably while strengthening security and controlling costs. Expected to enter public preview soon, Business Ontology is part of Snowflake Horizon Context, a data management layer within the company’s Horizon Catalog platform. The issue here is that AI agents can access vast volumes of data but may interpret that information incorrectly. Business Ontology addresses this challenge by enabling companies to establish official definitions for business terms, designate who owns those definitions and connect them to the data used in calculations. These approved definitions can then be shared across AI agents and business departments. As Snowflake touts it, the approach will reduce the development work required to deploy agents across large organizations. Rather than programming the same business rules into multiple applications, developers can use definitions maintained in a central repository. Three-Layer AI Context System Business Ontology operates alongside two other Snowflake capabilities: Cortex Sense and Semantic Views. Each performs a different function in helping AI systems interpret business information. Cortex Sense examines metadata, query histories and other data to identify relationships within enterprise data. Business Ontology provides human oversight by allowing authorized employees to approve or modify those interpretations. Semantic Views handles the calculations behind business metrics, providing a consistent method for querying data. Added together, the three capabilities are designed to lower the error rate caused by conflicting business definitions. Snowflake’s Horizon Context can collect metadata from enterprise applications including Tableau, Power BI, dbt and PostgreSQL. Business Ontology also supports open formats including Apache Ossie and Resource Description Framework, so definitions can be exchanged with other systems. This interoperability is a key competitive feature as Snowflake vies with Microsoft Fabric, Databricks and AWS for market share in enterprise AI deployments. Snowflake is also expanding Cortex AI Gateway, which provides controls over agent activity. New features include Cortex Sandboxes, which allow agents to execute code in isolated environments, and Agent Trace Intelligence, which helps developers examine agent operations. Another feature, Agent Abuse Detection, is built to identify agents attempting unauthorized actions or accessing restricted information. It’s currently in private preview. Snowflake is also expanding its CoCo agent development tools. CoCo Cloud Agents allows developers to run coding agents in managed cloud environments, while CoCo Automations supports scheduled workflows. Both are generally available. Given all the new releases, it’s likely that Snowflake’s new features will require further testing in large production environments. Capabilities as complex as the company’s new agent security and governance features will need to demonstrate sufficient maturity at scale. Another challenge is maintaining accurate business definitions as companies change their products, pricing and reporting practices. Different departments may also require different definitions of the same metric. In any case, the tools are designed to support return on investment for enterprise AI, which has been a challenge for companies. According to Deloitte’s 2026 State of AI in the Enterprise report, 28% of surveyed organizations reported revenue gains from AI, compared with 20% in January. Despite that improvement, most organizations had not yet reported higher revenue.