Snowflake launched Cortex AI Gateway on July 28, 2026, a centralized control layer designed to give enterprises visibility and control over their AI agent fleets. The product addresses two problems that are quietly becoming urgent as organizations scale beyond pilot deployments: nobody knows exactly what their AI agents are doing at any given moment, and the token bills are coming in larger than anyone budgeted for.

What Cortex AI Gateway Does

The gateway sits between an enterprise’s AI agents and the systems they reach into. When an agent calls a tool, queries data, or takes an action, the gateway logs it. When a model processes a request, the gateway captures the token count and attributes it to the specific agent, team, and workload that generated it.

The result is a complete audit trail. Enterprises can see which agent accessed which system, in what sequence, and what it cost. Finance and IT teams can then set spending limits that actually enforce before a bill arrives rather than discovering overruns after the fact.

Snowflake built the product directly on its acquisition of Natoma Labs, a startup that built enterprise-grade Model Context Protocol (MCP) infrastructure. MCP has become the standard way AI agents connect to external tools and data sources, and Natoma’s technology gives Cortex AI Gateway a principled way to broker those connections securely across an organization’s entire agent fleet.

Partners and Compatibility

The gateway does not require an all-Snowflake environment. It is designed to govern agents built inside Snowflake’s own platform alongside third-party agents built on external tools including Claude Code and Cursor. Snowflake is positioning the gateway as infrastructure for the broader agentic enterprise, not just a feature for Snowflake customers.

To address identity and access management at the agent layer, Snowflake announced integrations with 1Password, Aembit, Linx Security, Okta XAA, SailPoint, and Saviynt at launch. These partnerships close one of the harder problems in enterprise agent deployment: AI agents need credentials to access systems, and those credentials need to be managed with the same discipline applied to human employees.

Why Governance Has Become the Urgent Problem

A year ago, enterprises were asking whether AI agents could do useful work. Today most have run enough pilots to be confident they can. The question has shifted to whether organizations can scale agents safely and economically.

The governance gap is real. AI agents make autonomous decisions, access sensitive systems, and consume compute at rates that can multiply unexpectedly as usage grows. Without a control layer, an organization running dozens of agents across finance, operations, and customer service has limited visibility into what those agents are actually doing and limited ability to catch a problem before it becomes expensive.

The cost dimension is equally pressing. Token costs are no longer trivial at enterprise scale. An agent fleet handling thousands of daily transactions across multiple models can generate meaningful infrastructure spend, and attributing that spend to specific business functions is difficult without instrumentation purpose-built for the task.

Cortex AI Gateway is Snowflake’s answer to both problems. It positions Snowflake’s data cloud as the observation point for the enterprise agent layer, the place where organizations can see, govern, and account for the AI workforce they are building.

What This Means for Business

The practical implication of this launch is that enterprise AI agent governance is maturing from a nice-to-have into an infrastructure category with proper tooling.

If your organization is running AI agents in production, whether built on Snowflake or elsewhere, three things follow from where the market is heading.

Audit trails are no longer optional. Regulatory pressure from the EU AI Act and emerging US state requirements is moving toward mandatory logging for high-impact AI decisions. Organizations that build audit capability in from the start will have an easier path to compliance than those that add it later.

Agent costs need to be attributed, not pooled. The organizations getting the most out of AI agents are treating them like employees with budgets, not utilities with flat rates. Understanding which agents drive value and which consume resources without clear return is how you make good decisions about where to scale.

MCP governance is becoming the underlying standard. Snowflake’s decision to build Cortex AI Gateway on top of MCP infrastructure reflects where the industry is coalescing. If you are evaluating agent platforms, the ones building on open standards rather than proprietary connectors will be easier to govern and integrate over time.

The agentic enterprise is not a future state. It is what enterprises are building right now. The tools to govern that build-out are finally arriving.

---

Enterprise DNA designs and deploys AI agent systems for businesses across operations, data, and custom applications. If you are working through how to govern an agent fleet at scale, book a discovery call to talk through what the right architecture looks like for your business.