Snowflake is expanding what AI agents can do inside the enterprise, from investigating broken data pipelines to running recurring business analysis. Announced November 3 at Expedition 2026, the updates pair broader automation with controls over the definitions agents use, the systems they access and the money they spend.
The announcements bring together three operational concerns: whether an agent understands a business metric correctly, whether it has permission to act and whether the data feeding its work is reliable. Some capabilities are generally available. Others, including a proposed generation of always-on bots, remain in preview or have yet to enter it.
Controlling what agents mean and what they can do
Snowflake’s new Business Ontology is designed to give agents approved definitions of business concepts such as net revenue. Part of Horizon Context within Horizon Catalog, it connects those definitions to underlying tables and views, assigns ownership and records changes.
For teams using multiple analytics tools and agents, the practical issue is consistency. A revenue question should use the definition approved by the business, with a record of who approved it. Business Ontology is scheduled to enter public preview.
Cortex AI Gateway, first announced in July and now generally available, governs agent connections and provides activity records and spending controls. Its cost governance capabilities attribute usage and enforce budget limits.
Availability varies across the expanded gateway. MCP Tools is in public preview, while Agent Abuse Detection, Agent Trace Intelligence and Dynamic Model Routing are in private preview. These features should not be treated as a single production-ready package.
Agents investigate pipeline failures. Engineers approve the repair.
Snowflake is also putting CoCo, its coding agent for data teams, to work on recurring pipeline maintenance. Generally available CoCo Cloud Agents run in isolated environments managed by Snowflake, while CoCo Automations schedules tasks such as health checks and failed-job investigations.
When a scheduled investigation finds a problem, CoCo can trace the likely cause and propose a fix. The engineer decides whether that change enters production. The CoCo GitHub Integration, in public preview, carries proposed changes into the team’s existing code review process.
This gives data teams a way to delegate initial investigation and repair preparation while retaining responsibility for production changes. Pipeline Builder, also in public preview, adds a visual editor for connecting pipeline steps in Snowflake Workspaces.
Always-on bots are the next step, not today’s release
Snowflake’s planned CoCo and CoWork Bots would extend automation beyond predefined recurring jobs. Users would assign a goal, such as preparing customer meeting briefings or maintaining pipeline health, and the bots would plan and carry out work in the background.
Snowflake says the bots would inherit users’ permissions and return when a critical decision or final approval is needed. They are listed as coming to private preview, with no general availability date announced.
Scheduled automation is already available. CoCo Automations records task histories and allows teams to suspend or rerun jobs. CoWork supports recurring business analysis, including weekly sales reports, and its generally available Async Agent API supports longer tasks that continue in the background and resume after interruption.
Sources
ERP News Editorial Team
The ERPNews Editorial Team covers global developments in ERP (Enterprise Resource Planning), enterprise software, cloud platforms, AI, automation, and digital transformation, providing independent news and editorial analysis for senior business and technology leaders. Our reporting focuses on market signals, strategic shifts, and enterprise impact across the ERP and enterprise technology ecosystem.
For editorial inquiries, please contact:
📩 [email protected]
