Snowflake on April 21, 2026 expanded its Snowflake Intelligence and Cortex Code platforms with a set of features that turn the Bozeman-based data company into something it has never been before: a coordination layer for AI agents working across data its customers store somewhere else. The expansion adds native connectors to Gmail, Google Calendar, Google Docs, Jira, Salesforce and Slack, and extends Cortex Code to read directly from AWS Glue, Databricks and open-source PostgreSQL without forcing data migration. The strategic concession buried in the release is that Snowflake no longer expects to own all the data its agents touch.
The announcement, detailed in Snowflake’s April 21 press release, frames the moves as building a control plane for what the company calls the agentic enterprise. The label is marketing. The underlying shift is real.
What Snowflake Actually Shipped on April 21
Three product surfaces moved at once.
Snowflake Intelligence, the company’s business-user agent, gained Skills, Artifacts and a deep research mode. Skills let a user describe a recurring workflow in plain English, such as preparing a weekly pipeline review or drafting follow-up emails after a sales call, and have the agent execute it. Artifacts save and share analyses, dashboards and workflows so a colleague’s one-off report becomes reusable.
Deep research runs multi-step, fully cited reports across structured tables, unstructured documents and external context. An iOS app moved into preview alongside it.
Cortex Code, Snowflake’s builder layer, picked up support for AWS Glue, Databricks and PostgreSQL as data sources, plus a Visual Studio Code extension in private preview, a Claude Code plugin, and a Python and TypeScript Agent SDK. A new Cloud Agents capability includes Plan Mode and a Snap and Ask feature for screen-grab queries.
The connectors run on the Model Context Protocol and the Agent Communication Protocol, the two open standards that have become the de facto plumbing for cross-vendor agent coordination as of April 2026.
Why Snowflake Is Now Reading From Databricks Tables
The most consequential line in the release is the one that says Cortex Code supports Databricks, Postgres and AWS Glue as first-class sources.
For a decade, Snowflake’s pitch was the opposite: bring all your data here, query it together, pay for the compute. The new posture concedes that large enterprises run multiple data platforms in parallel and are not consolidating fast enough for an AI agent road map to wait. By treating rival platforms as legitimate read targets, Snowflake is betting it can win the orchestration layer even where it loses the storage layer.
The bet has competitive logic. Databricks crossed roughly $5 billion in annualized revenue in late 2025, drawing level with Snowflake on top-line scale while keeping a steeper growth curve and a higher private valuation. Conceding storage neutrality is cheaper than losing the agent layer outright.
Baris Gultekin, vice president of AI at Snowflake, framed the platform thesis directly. “AI is changing how every company operates, and the platforms that win will make it easy to put AI into practice with the right data and guardrails,” Gultekin said in the company’s announcement.
The Adoption Numbers Hidden Inside the Release
Snowflake disclosed a set of usage figures that read better than its competitors have managed.
- 9,100+ customers are using Snowflake AI products on a weekly basis as of the April release.
- 13,300+ total customers globally, putting active AI-product usage at roughly 68 percent of the installed base.
- 50%+ of customers have actively used Cortex Code since its launch in November 2025, a five-month adoption arc.
- 740 net new customers were added in the company’s most recent reported quarter, up about 40 percent year over year per Morningstar’s post-earnings analysis.
Morningstar lifted its fair-value estimate on the stock to $223 from $193 after the most recent quarter, citing AI-induced demand. The April 21 announcement is the first product proof point inside that revised thesis.
Skills and Artifacts Try to Crack the Reuse Problem
The more under-covered features in the release are aimed at a basic enterprise pain: knowledge work that gets done once and never gets done again.
Skills package a workflow as a reusable, named asset. A finance team that builds a quarterly variance review can save the entire chain, governance and all, and let any analyst rerun it. Artifacts do the same for outputs, turning a chart, a memo, or a multi-step analysis into something a teammate can fork.
Kumar Maddali, vice president of product development at Telenav, said the conversational layer has compressed elapsed time on routine analytics. “Processes taking weeks now take minutes through conversational self-service,” Maddali said in the announcement.
The stakes for that promise are visible in primary data. A Modern Data Company survey of 540 enterprise practitioners, cited in industry coverage of the launch, found 89 percent struggle to find relevant data inside their own organizations and 60 percent lack any internal data discovery tool. Skills and Artifacts are aimed precisely at that gap.
How the Cortex Code Builder Layer Stacks Up
Against Databricks
Databricks responded earlier in 2026 by adding SQL-based AI document parsing inside its Agent Bricks framework, mirroring a Snowflake feature within days of release. InfoWorld’s reporting on the agent build-out noted that the two companies’ feature lists have converged to the point that procurement decisions now turn on ease of use versus open-source flexibility rather than raw capability.
Against the Hyperscalers
The VS Code extension, the Claude Code plugin and the Python and TypeScript SDKs go after a different audience: working developers who already chose AWS, Azure or Google Cloud for compute. By meeting them inside their existing IDE, Snowflake avoids forcing a tool switch.
The Governance Argument Snowflake Is Quietly Making
The most defensible piece of Snowflake’s pitch is governance.
Every connector, every Skill, every Artifact inherits the same role-based access controls, lineage and audit logs the underlying Snowflake account already enforces. That matters because an AI agent that can read a customer record from Salesforce, write a follow-up in Gmail, file a ticket in Jira, and update a forecast in a Snowflake table is also an agent that can leak across all four if its permissions are wrong.
“Snowflake Intelligence accelerates decision-making, reducing overhead for services,” said Sameer Vuyyuru, chief AI and product officer at Capita.
Wolfspeed has pushed the model further inside operations. Priya Almelkar, the chip-maker’s chief information officer, said in the release that the company has “deployed dozens of AI agents across manufacturing, quality, supply chain.” Tony Leopold, chief technology and strategy officer at United Rentals, said teams now “access real-time insights without analysts using natural language.”
What This Does to the SNOW Investment Case
The product side of the announcement reinforces the bull thesis already priced into the stock since the last earnings beat. The strategic side introduces a new variable.
If Snowflake wins the control-plane bet, future revenue stops being a function of how much data sits in Snowflake and starts being a function of how many agent calls flow through Snowflake regardless of where the data lives. That is a larger addressable surface. It is also a more contested one, because every hyperscaler and every model vendor wants the same job.
The company has scheduled additional rollout milestones through the second half of 2026, with deep research and Artifacts moving from preview to general availability. The Cortex Code support for external systems is live now.
Frequently Asked Questions
What is Snowflake Intelligence?
Snowflake Intelligence is the company’s business-user AI agent, launched in earlier preview and expanded on April 21, 2026 with Skills, Artifacts, deep research and an iOS app. It works on top of governed data already inside a customer’s Snowflake account and now connects out to Gmail, Google Calendar, Google Docs, Jira, Salesforce and Slack.
What is Cortex Code?
Cortex Code is Snowflake’s builder layer for developers, first launched in November 2025. As of April 2026 it supports AWS Glue, Databricks and PostgreSQL as data sources, ships a VS Code extension in private preview, integrates with Claude Code, and provides a Python and TypeScript Agent SDK.
How is Snowflake competing with Databricks on AI agents?
Snowflake is positioning itself as the cross-platform control plane while Databricks pushes deeper agent capabilities inside its own lakehouse. The two companies have reached near-parity on individual features, with procurement decisions now turning on ease of use, open-source flexibility and governance posture rather than raw functionality.
What does control plane mean in this context?
A control plane is the orchestration layer that decides which agent runs, on which data, with which permissions, and using which model. By owning the control plane, a vendor captures usage value even when the underlying data and the underlying model live elsewhere.
Are the new features available now?
Cortex Code’s expanded external-system support is live as of April 21, 2026. Skills, Artifacts, deep research and the iOS app are in preview, with general availability expected later in 2026 according to the company’s release.
The reach of the bet shows up in one number Snowflake did not put in a chart. Roughly 9,100 of its 13,300 customers now touch an AI feature every week, and the company is wagering that the next billion dollars of growth comes from agents calling data that Snowflake never owned in the first place. Whether that bet pays will be visible by the time the company reports its next two quarters.
Leave a Comment