Anyone who has attended an enterprise technology conference over the past year could be forgiven for reaching the same conclusion: AI cannot operate at scale – or deliver predictable business outcomes – without a unified data foundation. Other concerns like security, governance, cost, control, and context are also key concerns.
Natoma’s role
Snowflake is doubling down on security and intelligent model orchestration as it weaves its recent Natoma acquisition into the fabric of its AI strategy.
Natoma is a centralised model context protocol (MCP) gateway that enforces identity, policy, and audit at the tool-call level for AI agents.
Framed as extending Snowflake’s governed database, Snowflake has already folded Natoma’s gateway capabilities into a new product called Cortex AI Gateway. This gateway is meant to be a trusted control plane that governs how AI agents securely access AI models, tools, MCP servers, enterprise systems, and data.
It is also intelligent enough to route agent requests to the appropriate model based on cost, latency, capability and data residency.
Snowflake’s no model lock-in stance combined with intelligent model routing, is also positioned as a major cost optimisation measure.
“You will typically be seeing anywhere from two to eight times cost savings when you have a good model routing layer,” said Snowflake’s senior vice president of customer experience and frontier engineering, Madhan Arumugam.
Systems of record
Farhan Choudhary, Snowflake’s senior product marketing manager said, “We want integrations with all major systems of record,” adding that Snowflake has AI-powered migration accelerators and that MCPs currently are one of the ways to integrate with the data “very securely”.
“Our acquisition of Natoma – once we’ve integrated all of these capabilities – the security and governance will be permeated into that layer as well.”
We want integrations with all major systems of record.
Farhan Chaudhary
The comment underlines Snowflake’s intent to use Natoma to police data flows and enforce policies across agentic AI workloads, an emphasis echoed publicly by Snowflake’s chief security and trust officer, Mayank Upadhyay, who has said the company wants security to be at the centre of enterprise AI’s shift from data interoperability to agent interoperability.
The agentic control plane
Snowflake calls CoCo (formerly Cortex Code) and CoWork (a personal agent formerly called Snowflake Intelligence) its “agentic control plane” because the company positions the pair not as chatbots but as the coordination layer sitting between AI agents and the rest of the enterprise.
Executive vice president of product Christian Kleinerman made the framing explicit at Snowflake Summit 2026 earlier this year, telling attendees that “all of this is coordinated by what we call an agentic control plane.”
CEO Sridhar Ramaswamy has described the two products as “the first major product surfaces of the agentic control plane” which suggests that more products and solutions for Snowflake’s agentic control plane are in the pipeline.
This was re-emphasised by Madhan during the KL edition of the Snowflake World Data Tour who even labelled them as “habit-forming agents” and the “two primary interfaces” to an enterprise’s data.
According to the Shridar, both CoCo and CoWork are expected to “…coordinate across your context, models and applications, so that decisions and actions happen seamlessly across your business.”
