Snowflake expands AI governance tools to move enterprise agents into production

Snowflake has introduced new tools to help businesses deploy AI agents with shared business context, centralized security controls and autonomous workflows. The updates aim to address challenges that prevent enterprise AI projects from reaching production.

(For more news, click here)

Businesses are moving beyond AI assistants that answer questions toward agents that can query systems, update records, and carry out recurring tasks. That shift brings a practical problem: software cannot safely act on company information unless it understands how the business defines that information and what actions it is permitted to take.

Snowflake has announced a series of updates aimed at that transition, expanding its Horizon Catalog, Cortex AI Gateway, CoCo and CoWork products. Introduced during Snowflake Expedition 2026, the additions are designed to give AI agents approved business definitions, centralized access controls, and tools for executing work with less direct supervision.

The challenge extends beyond technical integration. In a Snowflake-sponsored MIT Technology Review Insights survey of 400 senior technology executives, 55% cited insufficient business context as the leading reason agents produced poor-quality insights, while 45% pointed to fragmented or inaccessible data.

Giving AI agents consistent business context

Snowflake’s Business Ontology feature is intended to establish shared definitions for terms that can mean different things across departments. A request for net revenue, for example, may produce inconsistent results when finance, sales and analytics teams apply different calculations. Through Horizon Catalog, organizations would be able to approve definitions, assign responsible owners and connect terms to underlying data, including sources outside Snowflake.

The system works alongside Semantic Views, which calculate defined metrics, and Cortex Sense, which can infer context that has not been formally documented. Snowflake says approved definitions will take precedence over inferred ones, while changes can be logged for auditing.

Existing definitions may also be imported from spreadsheets or third-party catalogs using CoCo. Business Ontology is approaching public preview, while Cortex Sense remains in private preview.

Managing permissions, security and AI costs

Correctly interpreting data is only part of the challenge when an agent can access business applications or trigger actions. Cortex AI Gateway is designed to apply centralized policies across models, tools and cloud environments, including connections made through Model Context Protocol (MCP) servers. This allows companies to use established access permissions rather than creating separate controls for every agent.

The security implications extend beyond individual platforms. Tech Revolt previously reported that the growing use of autonomous agents is creating accountability challenges as organizations grant software greater access to business workflows.

Snowflake is addressing some of these concerns through tool-call authorization, protections against prompt injection, and controls for tracking AI spending by department.

Other capabilities include Dynamic Model Routing, which directs routine requests to less expensive models, and Agent Abuse Detection, which examines sequences of actions for suspicious behavior. The gateway is generally available, but several surrounding features remain in preview.

Christian Kleinerman, Snowflake’s Executive Vice President of Product, said, “The enterprises that win won’t be the ones with the most agents.” He emphasized the importance of agents understanding approved business definitions, following governance requirements and accounting for their actions.

From recurring tasks to autonomous workflows

Snowflake is also extending CoCo and CoWork, its coding agent for data teams and assistant for business users. CoCo Automations can run scheduled activities such as checking data pipelines and preparing recurring reports, retaining execution histories that teams can inspect or rerun. Integrations with Slack, Excel and Google Sheets are intended to make these capabilities accessible through applications employees already use.

The company also outlined CoCo and CoWork Bots, which would work toward broader goals using an employee’s existing permissions and request approval when needed. Unlike the available automation features, these bots remain in private preview.

Snowflake highlighted customers in the UAE and Saudi Arabia among organizations using its broader data and AI technologies, although it did not confirm that all had deployed the newly announced agents.