Snowflake wants employees to stop chasing AI agents for answers and start delegating entire jobs to them.
The company announced always-on bots for CoCo, its coding agent for data teams, and CoWork, its personal work agent for knowledge workers, at Snowflake Expedition on Nov. 3. Coming soon to private preview, the bots are designed to plan and complete assignments in the background, requesting human decisions or approval when needed.
Users can give the bots broader objectives, such as preparing customer meeting briefings or monitoring data pipelines. The agents then organize the necessary steps and execute them using the user’s existing access rights, role context and permissions.
Snowflake reported more than 9,100 accounts using CoCo and more than 5,800 using CoWork, based on average weekly usage during the final four weeks of the quarter ended July 31, 2026. The announcement follows other enterprise automation developments, including Omnissa’s Elara agents for IT management. CoCo Automations is generally available in Snowsight and the command-line interface, with public preview in CoCo Desktop. It allows data teams to schedule recurring workflows, including pipeline health checks and routine reporting.
Each execution maintains an audit trail and task history, giving teams visibility into completed work and the ability to suspend or rerun jobs. CoWork can similarly schedule recurring business analysis, while its Async Agent API supports longer tasks that continue running in the background and resume after interruptions.
Snowflake is also bringing its agents into everyday business applications. Snowflake for Slack, CoCo for Excel and Sheets, and Snowflake Mobile are in public preview. These integrations let employees access enterprise data, ask questions and manage agent activities without constantly switching between applications.
Governance becomes central to agent deployment
Allowing agents to operate independently creates security and cost concerns, particularly when they can access sensitive business information or execute actions across enterprise systems.
Snowflake is addressing these concerns through its broader governance infrastructure. The company describes Cortex AI Gateway as a central control point for monitoring agent activity, governing access to tools and models, and managing AI spending. Partners should confirm which controls are available for each proposed deployment.
The company also announced Business Ontology, coming soon to public preview, to give agents consistent, approved definitions of business terms. This matters because departments can interpret the same metric differently, potentially producing conflicting answers from otherwise capable AI systems. Other additions include shared workflows, role-based profiles and memory features that let teams retain useful context while controlling what agents remember. Agent oversight is also a focus elsewhere in the market, including Workday’s Agent System of Record.
What this means for enterprise technology partners
Snowflake’s announcement creates opportunities for channel partners, particularly systems integrators, managed service providers and data consultants helping enterprises move AI projects beyond experimentation.
Customers will need more than agent licenses to deploy autonomous workflows successfully. They will need assistance connecting enterprise data, establishing permissions, integrating existing applications and setting up monitoring and approval processes.
Partners could build services around these requirements, including agent deployment, governance assessments, workflow automation and ongoing optimization. Snowflake’s support for multiple models also creates room for partners to help customers select models based on workload complexity and operating costs.
However, the new bots are still awaiting private preview. Partners should distinguish between capabilities announced for future testing and features customers can deploy today.
What partners should establish before deployment
Partners should start with a narrowly defined workflow, such as preparing a weekly sales report or checking pipeline health. Before testing, they should identify the required data sources, permissions and approval points, along with the employee responsible for reviewing results.
A pilot should measure completed work, correction rates and operating costs. Partners should also test how the workflow responds when data is missing, a connected tool fails or an agent requests an action outside its intended scope. Those results can help customers decide whether to expand the deployment.
Ongoing services could include reviewing execution logs, maintaining integrations, updating approved business definitions and adjusting spending limits. Partners should establish who investigates failed jobs, approves changes and pauses an automation when its output becomes unreliable.
Customer proposals should distinguish generally available capabilities from previews and forthcoming features. With the new bots still awaiting private preview, partners can assess workflow readiness now, while making deployment commitments conditional on confirmed access, supported functionality and commercial terms.
Read more: Explore why enterprise AI agent adoption hinges on guardrails and how partners can help customers establish approval workflows, audit trails and operational controls.