Snowflake is expanding two AI products for different enterprise audiences. Snowflake Intelligence gives general business users natural-language access to data and workflows, while Cortex Code provides developers with coding, orchestration and agent-building tools.
The strategy keeps both groups close to Snowflake’s governed data platform. New connectors, web interfaces and software-development kits aim to make agents easier to use without bypassing the permissions already attached to enterprise information.
Snowflake says more than 9,100 customers use its AI products weekly and that over half its customers have adopted Snowflake Intelligence or Cortex Code within six months of launch.
Business users gain agents inside familiar workflows
Snowflake Intelligence can prepare presentations, conduct multi-step analysis and send follow-up messages from natural-language instructions. It works with structured and unstructured internal data alongside connected external systems.
New Model Context Protocol integrations include Google’s business suite, Jira and Salesforce products including Slack. An iOS application is expected to enter public preview, bringing the interface closer to mobile work.
The platform personalises over time and lets users save and share workflows. Longer context means employees should not need to repeat extensive background instructions for recurring tasks.
Personalisation increases utility but also requires clear boundaries. Saved context may contain sensitive assumptions or information that should not transfer to another user. Snowflake says queries and workflows remain constrained by organisational governance and access permissions.
Cortex Code targets enterprise developers
Cortex Code is positioned as a coding and orchestration layer. It can connect with AWS Glue, Databricks and Postgres, while MCP and Agent Communication Protocol support opens routes to external models and services.
A Visual Studio Code extension is in private preview, and Snowflake is developing a plugin for Claude Code. Python and TypeScript agent SDKs let teams embed Cortex capabilities into their own applications.
Cloud Agents will appear in Snowflake’s browser-based Snowsight interface. Plan Mode lets users inspect and approve a proposed workflow before execution, addressing the risk of an agent taking an unexpected route through data or code.
Snowflake also plans to expose more detail about longer research processes so users can assess how a result was produced. That visibility matters because a polished answer alone cannot reveal whether an agent used incomplete sources or made an unsupported inference.
One platform can reduce fragmented governance
Enterprises often adopt separate assistants for business analysis and software development. Each creates another connector, policy layer and audit trail. Snowflake’s dual approach offers common data access and governance while tailoring interfaces to different skills.
The benefit depends on whether controls remain consistent across integrations. An employee should not gain broader access simply because an agent reaches Salesforce through MCP rather than a traditional connector. Tool calls, shared workflows and personalised context all need traceable identities.
Snowflake developed parts of the update from Project SnowWork, a research initiative used to collect customer preferences. That feedback-led process may help the company prioritise practical workflow gaps instead of adding generic agent features.
The platform approach also creates dependence. Customers should understand how easily workflows and logs can move if they use another model or data environment. Support for multiple protocols and external models can preserve options if implemented beyond nominal compatibility.
Snowflake is widening its audience without abandoning its technical base. Business users receive an accessible agent, while developers receive tools to build and integrate governed systems.
The common requirement is trust in the data and action path. If Snowflake can make permissions, plans and provenance visible across both products, it may help organisations turn isolated AI experiments into reusable workflows without creating a parallel, ungoverned application estate. That visibility will decide whether convenience becomes dependable enterprise practice.
