If you work with enterprise data, Snowflake’s April 21 announcement is worth your attention. The company expanded two of its core AI products, Snowflake Intelligence and Cortex Code, and made clear what it is actually building toward: a single control plane for everything that happens in an AI-driven organisation.

That is a big claim. Here is what it means in practice.

Snowflake Intelligence Becomes a Personal Work Agent

The headline feature is that Snowflake Intelligence is no longer just a search and analysis tool. It now behaves like a personal AI agent that adapts to individual users over time, learning preferences, remembering workflows, and delivering more relevant outputs the more you use it.

More substantively, Snowflake added what it calls deep research to Intelligence, now in public preview. This is an agentic capability that lets users ask complex questions and receive fully cited, multi-step reports. The underlying architecture reasons across structured data (your database tables and warehouse), unstructured content (documents, reports, emails), and external context from the web, returning a single coherent answer with sources attached, not a list of raw results you still have to synthesise yourself.

For data teams, this is meaningful. One of the persistent frustrations with enterprise AI tools is that they are either good at querying structured data or good at reasoning over documents, but rarely both at once. Deep research is Snowflake’s attempt to close that gap at the application layer.

Cortex Code: Now Connecting Data Everywhere

Cortex Code is Snowflake’s AI coding assistant built for data-centric workflows. More than half of Snowflake’s customers have used it since its launch in late 2025.

The April 21 update extends Cortex Code outside the Snowflake environment. Developers can now build applications that pull from AWS Glue, Databricks, and PostgreSQL without migrating any data into Snowflake first. The tool works across those external systems through new integrations, and connects to enterprise applications using the Model Context Protocol (MCP), the same open standard that Anthropic introduced to let AI tools interact with third-party services in a standardised way.

For teams running a mixed data estate, which is almost everyone in enterprise today, this matters. You no longer need everything centralised in Snowflake to build AI workflows on top of it.

The update also ships with a VS Code plugin and a Claude Code plugin, putting Cortex Code directly into the environments where developers already work.

The “Control Plane” Play

The phrase Snowflake keeps using is “control plane for the agentic enterprise.” It is a deliberate positioning statement, one that puts Snowflake in competition with every other platform claiming to be the place where enterprise AI agents live, coordinate, and execute work.

The bet makes sense from where Snowflake sits. If your organisation’s data is in Snowflake, and your AI agents need access to that data to act on it, then whoever controls the data layer has a legitimate claim to also control the coordination layer. Snowflake is essentially arguing that the data warehouse is the right foundation for agentic systems, rather than the CRM (Salesforce’s argument), the workflow automation layer (ServiceNow and UiPath), or the foundational model (Anthropic, OpenAI, Google).

It is the same argument that won the cloud data warehouse era. Whether it wins the agentic era is a different question.

What This Means for Business

If your data is in Snowflake, your options just got better. Deep research and the new cross-platform integrations significantly increase what you can build without switching tools or migrating data. Teams using Snowflake for analytics who have been waiting for agentic capabilities now have more to work with.

The “no migration required” point is real. Connecting Cortex Code to AWS Glue, Databricks, and PostgreSQL without moving data is a practical unlock. Data migration is expensive, slow, and disruptive. Tools that work with your data where it lives lower the activation energy for AI adoption considerably.

The personal work agent framing is where this is all heading. Snowflake, Salesforce, Microsoft, and Google are all converging on the same vision: an AI layer that learns your individual workflows and automates routine work. The platforms are differentiated by which data they sit on top of, not the concept. The question for any business is which platform already has the data your agents need to be useful.

For organisations serious about building an agentic data infrastructure, Snowflake’s April 21 update is a step toward a more complete picture. The deep research capability in particular will get used by data teams the moment they have access to it.