[San Francisco, United States = Chi-gyu Hwang] Snowflake made clear at its annual Snowflake Summit 2026 conference that it plans to expand its role beyond a data storage and analytics platform to serve as a gateway for companies using AI agents.

This has put Salesforce into head-to-head competition in the enterprise AI agent platform race with companies it previously did not face as often, such as SAP and Salesforce. Most are companies with a voice in business applications. At the same time, model developers such as Anthropic and OpenAI are rapidly moving into enterprise AI agent services.

Snowflake said its strategy that starts from a data platform has competitiveness of its own.

Snowflake CEO Sridhar Ramaswamy (스리다르 라마스와미) and product chief vice president Christian Kleinerman (크리스티안 클레이너만) held a media and analyst briefing at the Snowflake Summit 2026 venue on June 1 local time. They described the current competitive landscape along three axes.

Business application companies such as SAP and Salesforce have strengths in work context, including sales processes and approval workflows.

Model providers may have an advantage in a new-player mindset that can innovate freely without being tied to existing structures.

What about Snowflake? Ramaswamy pointed to data, data gravity and a starting point in the enterprise context as differentiators. He said applications and services will naturally come to where a lot of data has accumulated.

"It is not possible to know at this point who will win this competition. History makes it look like the winner was decided only after the conclusion," he said. "Snowflake is operating Snowflake Intelligence and Coco, and accumulating data on actual usage patterns and failure points. We can directly see which skills users get stuck on and where they do not even try," he said. He said a platform that knows enterprise data most deeply could have an advantage in the agent race.

Kleinerman said entrusting the area to outsiders is a major strategic risk at a time when agentic interfaces are redefining how users interact with software. He said running in-house products makes it possible to secure telemetry data directly on where users get stuck and which functions they cannot use, and that this directly links to product improvement and customer support.

Ramaswamy acknowledged that all software companies, including Snowflake, need to be on guard as Anthropic and OpenAI each grow into companies worth $1 trillion.

Even so, he expressed confidence that Snowflake does not rely on revenue from closed models as open-source model innovation accelerates. He said model independence puts it in a favorable position to rapidly lower customer costs, and that model independence will be a factor that provides real value to enterprises.

On efforts to rationalise AI costs, he said a method of tiering frontier models and open-source models by type of work is spreading.

"You do not need to use a large model for the task of classifying customer feedback into 10 categories," he said. "On the other hand, for coding or complex planning tasks, top-tier model reasoning ability still makes a meaningful difference," he said.

Snowflake said it also developed a way, for its coding AI agent Coco (Cortex Code), to compile simple skills into deterministic code and run them without calling a model. It is part of an effort to cut token costs by calling models only for complex questions and handling simple repetitive tasks with general computation.

On Coco's impact on data engineering work, it stressed role shifts rather than headcount reductions. "Data engineers and data scientists are using Coco to create specialised agents that departments such as HR, finance and investor relations can use, and are taking on the role of spreading AI across the company," Ramaswamy said. "Data functions will change significantly, but the same workforce will have higher standing as the main players leading AI innovation within companies," he said.