Snowflake adds business context and controls for enterprise AI agents

Snowflake says Business Ontology is a capability within Horizon Context, the governed context layer in Horizon Catalog.[3] It lets teams define, own and approve shared business terms, entities, relationships and metrics, then connect those definitions to the data and models that implement them.[3] The capability records who owns a definition and whether it is approved, and links it to the semantic views, tables and views that implement it.[3] Snowflake said it is coming to public preview soon.[3]

The ontology is intended to clarify what a business term means, who owns it and how it relates to other concepts.[3] Snowflake says it preserves intentional differences in definitions across domains such as sales, finance and marketing, rather than requiring one shared meaning for every team.[3] Imported items are marked as drafts and routed for owner approval through a steward workflow.[4] Snowflake cited a lack of business context as the primary cause of low-quality AI-agent insights for 55% of executives; fragmented or inaccessible data ranked second at 45%.[2]

Cortex AI Gateway applies consistent policies to agent connections across supported clouds, frameworks and AI providers.[4] Its Tools feature authenticates agent tool calls with the end user’s identity and checks each call against policy before it runs.[4] Snowflake says Guardrails protects agents routed through the gateway against prompt injection, jailbreaks and novel attacks.[4] Cost Governance attributes AI spending to teams, departments, agents or applications and can enforce hard limits.[4] Dynamic Model Routing selects an organization-approved model for each request based on cost, accuracy and latency.[4]

Snowflake says Dynamic Model Routing delivered up to three times greater token efficiency than a frontier-model-only approach at the same quality.[4] Its generally available Observability feature records model calls, tool invocations, reasoning steps and guardrail decisions in governed tables in customers’ Snowflake accounts.[4] CoCo Automations lets data teams configure recurring workflows for agents to execute autonomously.[5] CoCo and CoWork Bots can work in the background toward broader goals, seek human input when decisions or approvals are needed, and inherit users’ access rights and permissions.[1][5] Snowflake said it has no exact figure for cost savings or faster production because results depend on the company.[1]