On 3 November 2026, at its Expedition 2026 event, Snowflake announced Business Ontology, a governed store of approved business definitions for AI agents, alongside new Cortex AI Gateway controls for tool access, spending and tracing. Several parts are still in preview, and SiliconANGLE places the launch in a wider race to build control layers for enterprise agents.
What happened
- Business Ontology. Snowflake describes it as a place to record approved definitions of business terms, link them to tables and views, and serve only approved definitions to agents. Snowflake lists it as "public preview soon" and SiliconANGLE reports it is in private preview now, so customers cannot generally use it yet.
- Cortex AI Gateway. This is Snowflake's single entry point for agents, applying one set of access policies across clouds, frameworks and AI providers. SiliconANGLE reports that it was first introduced in July and expanded this week. Snowflake's own materials differ on its status: the press release says generally available, while two blog posts from the same day say public preview and the September release note says preview. Check the documentation before planning around it.
- Controls by status. Snowflake lists cost governance, including hard spending limits by team, agent or application, and the recording of agent activity in the customer's own account as generally available. Checking each agent tool call against policy (Cortex AI Gateway Tools) is in public preview. Cortex Sandboxes, Agent Abuse Detection, Agent Trace Intelligence and Dynamic Model Routing are in private preview.
- Always-on agents. Snowflake Bots, which Snowflake describes as goal-driven agents running tasks over days or weeks under existing permissions, are listed as "private preview soon".
- Wider market. SiliconANGLE also places the launch alongside similar work from SAP and Salesforce, and reports an expanded Oracle partnership that will offer Snowflake as a first-party service on Oracle Cloud Infrastructure.
- The survey behind the pitch. Snowflake's release cites an MIT Technology Review Insights report sponsored by Snowflake. It surveyed 400 senior data and technology executives at organisations earning $500 million or more a year, and 55% named a lack of business context as the main cause of low-quality agent insights. Treat this as vendor-sponsored research on large companies.
What it means for businesses
The useful signal is not one vendor's feature list. It is that three practical problems now sit at the centre of enterprise agent products: what your business terms mean, what an agent is allowed to touch, and how much it may spend.
Those problems exist whichever platform you use. If "active customer" or "net revenue" has three meanings across your systems, an agent can give a confident answer using the wrong one. If an agent can call a tool with broad service credentials, a mistake can reach a live system. If usage has no limit, a looping agent can run up costs before anyone notices.
Preview labels matter too. A feature in private preview is not something to base a project plan on, and vendor capability descriptions have not been independently tested here. These controls also only cover agents routed through the gateway, so systems outside it still need their own rules.
What to do next
- List the agents and AI tools already in use, including those inside SaaS products, and record what data and actions each one can reach.
- Write down your ten most used business terms, each with one owner and one formula. Our guide to business dashboard KPI design gives a definition template you can reuse.
- Give agents limited identities. Avoid shared admin credentials, and require human approval for actions that are hard to undo.
- Set a spending limit and an alert per team or per agent before scaling up usage.
- Keep logs you can query: the model calls, the tool calls and the identity behind each one.
- Run a small pilot with a measured outcome before you buy a platform. Our guide on how to evaluate an AI pilot sets out the checks.
- Ask every vendor for the status of each feature in writing: generally available, public preview or private preview.
Timeline Digital, a custom software development company founded in 2013 in Islamabad, builds AI agents as managed projects, with narrow tool permissions, human approval steps and audit logs designed around the client's own systems. Every project starts with a free pilot: Before you commit to the full project, we build 2 to 3 of your key modules as working software, free of charge, ready in about 2 days. Your team tests the pilot, and the full build starts only after you approve it.
Sources
- Snowflake Helps Enterprises Deploy AI Agents with Trusted Business Context and Built-In Controls, Snowflake press release, 3 November 2026
- Governed Context, Security and Spend Controls for AI Agents, Snowflake blog, 3 November 2026
- Build Faster, Work Smarter with AI Agents and Apps Grounded in Data, Snowflake blog, 3 November 2026
- Cortex AI Gateway (Preview) release note, Snowflake documentation, 15 September 2026
- Snowflake expands agentic AI platform with business context and governance tools, SiliconANGLE, 3 November 2026