Snowflake Pushes to Become the Control Plane for Agentic AI

Snowflake is no longer positioning Intelligence and Cortex Code as separate AI add-ons. It is describing them as the operating layer for what it calls the agentic enterprise, where agents do not merely summarize data but take actions across systems ^[1][3][6]. That framing matters because it moves the conversation from model selection to orchestration, permissions, and execution. In other words, Snowflake wants to sit where enterprise AI becomes operational.

The competitive backdrop is crowded. Analysts cited in the coverage pointed out that Databricks, AWS, Google Cloud and Microsoft are all racing to assemble similarly broad AI ecosystems, which makes many of the new capabilities feel like required investments rather than singular differentiators ^[3]. Snowflake's claim to distinctiveness is that it can unify data, governance and action in one perimeter, instead of asking customers to stitch those layers together themselves ^[4][6].

For nontechnical employees, the headline feature is Skills: a natural-language way to define repeatable work such as preparing presentations, running multistep analysis, or sending follow-ups ^[4][7][8]. Snowflake is also adding MCP connectors for Gmail, Google Calendar, Google Docs, Jira, Salesforce and Slack, which makes the assistant useful inside the tools people already use ^[3][6][7]. Deep research adds cited, multistep reporting, while Artifacts lets teams save and share analyses and workflows so work can be reused instead of recreated ^[6][7][8].

The iOS app extends that experience beyond the desktop, and that is not a trivial addition. In many enterprises, decisions happen in transit, on the floor, or between meetings, so mobile access can determine whether AI gets used for real work or remains a dashboard novelty ^[6][9]. Snowflake is trying to make data interaction feel less like querying a warehouse and more like delegating a governed task to a capable assistant ^[4][8].

Cortex Code is being expanded from a coding assistant into a broader builder layer for enterprise agents. Snowflake now supports external systems such as AWS Glue, Databricks and Postgres, so developers can work with data where it already lives instead of migrating everything first ^[3][6][7][8]. The platform also connects through MCP and the Agent Communication Protocol, which is a sign that Snowflake wants to coexist with other agent frameworks rather than force customers into a closed environment ^[3][6][9].

The developer experience is also moving closer to familiar tools. VS Code support, a Claude Code plugin, and Python and TypeScript SDKs are meant to lower the friction of embedding agent capabilities directly into existing software projects ^[6][7][8]. Cloud Agents in Snowsight, plus Plan Mode and Snap & Ask, suggest Snowflake understands that builders want preview, inspection and rollback before they let code touch production workflows ^[4][7][8].

The most consequential part of the announcement is not the UI polish or connector count. Snowflake is insisting that the same controls used for data access, auditing, row-level permissions and budget management should govern agent behavior too ^[4]. That is the practical answer to the fear that AI systems will act quickly but recklessly, pulling from the wrong sources, crossing role boundaries, or triggering expensive workflows without oversight. For enterprise buyers, trust is becoming a product feature.

That trust gap is real. TechTarget noted a July 2025 MIT report finding that 95% of organizations had not yet seen a return on AI investments, while a separate survey of 540 data practitioners found that 89% struggle to find the right data and many lack good discovery tooling ^[3][6]. Snowflake's argument is that agents fail less because models are weak than because the underlying data and workflow context are fragmented. If that diagnosis is right, the winning platforms will be the ones that reduce operational friction, not just generate better text.

Snowflake is already pointing to customers that treat the platform as operational infrastructure. United Rentals said more than 1,600 locations are using Snowflake Intelligence to access real-time insights, while Cortex Code is helping the company build and scale AI agents to improve sales growth and fleet availability ^[4]. Accenture said thousands of practitioners are using the platform across client accounts and nearly two dozen skills, spanning SQL development, notebooks and semantic modeling ^[8]. Those examples suggest the early payoff comes from compressing the time between a question, an analysis and an action.

Capita's use case hints at another pattern: fragmented contact-center data becomes far more valuable when agents can unify it into live, natural-language operations ^[8]. That model translates well to other distributed environments where teams depend on timely decisions and repeatable processes, from public-service operations to field fleets and manufacturing support ^[6][9]. Snowflake's challenge now is to prove that these workflows keep working when they leave pilot mode and hit production scale.