Snowflake launches CoWork & Cortex Training AI tools

Snowflake has introduced new artificial intelligence products under its CoWork and Cortex Training lines, expanding its push to embed AI tools in day-to-day business work.

The updates include new functions for knowledge workers through Snowflake CoWork, formerly Snowflake Intelligence, and a new model training offering called Cortex Training. Snowflake is positioning the products as a way for companies to connect internal data, business context and software tools in one governed environment.

New CoWork features include Artifacts, Cortex Sense, User Skills, User Memory and a Skill Catalog, along with integrations with Google Drive, Salesforce and Slack through Model Context Protocol connectors. Snowflake is also adding a CoWork iOS app, a Slack bot and a Microsoft Excel extension.

CoWork is aimed at knowledge workers dealing with fragmented data, disconnected tools and workflows that do not share context. Deep Research, another part of the product set, searches structured and unstructured company data through multi-step analysis and returns cited responses within the company's governed data environment.

Snowflake said it now has more than 13,900 customers worldwide and that more than 13,600 accounts use its AI products each week, underscoring how central AI has become to the company's broader data platform strategy.

Christian Kleinerman, EVP of Product at Snowflake, said the company sees a broader role for AI inside large organisations.

"The enterprises with real competitive advantage are the ones putting the full power of AI action in the hands of every single person in their organization," Kleinerman said.

"CoWork gives every employee a personal agent to work smarter across their business data and tools, so they can get insights, take action, and collaborate in one place, all within a secure, governed environment they trust. This is the future we're building, where AI doesn't just surface insights, but becomes the control plane that connects data, context, and enterprise systems to drive action across the business."

From insight to action

A central part of the launch is Snowflake's effort to shift business AI from answering questions to carrying out tasks. CoWork can automate workflows with User Skills and use User Memory to learn from individual behaviour, schedule tasks and provide recommendations through tools workers already use, including Gmail and Slack.

Cortex Sense is designed to provide a shared layer of business definitions, operational knowledge and data context for AI agents. Snowflake said this shared context is intended to reduce manual set-up and help companies deploy agents for functions such as finance and sales without rebuilding business logic for each one.

Artifacts now includes publishable dashboards built on live company data. Snowflake said analysts can create dashboards that business users can explore through natural language queries rather than standard clicks and filters alone.

Snowflake also linked the announcements to its plan to acquire Natoma, whose enterprise Model Context Protocol platform is intended to improve connectivity, governance, identity-aware authorisation and auditability for AI agents operating across business software.

Customer use cases

Snowflake cited Synopsys, Whoop and Under Armour among customers using CoWork. At Synopsys, internal teams have built more than 20 agents for functions including revenue operations, legal, finance, product management and IT, according to Snowflake.

"Snowflake has become the foundation for how our teams access, understand, and act on data across the business. We've built more than 20 purpose-built agents for teams spanning revenue operations, legal, finance, product management, and IT, and every one of them connects back to Snowflake because that's where our data, governance, and business context live," said Ramji Jagannathan, VP of Enterprise Digital Platforms at Synopsys.

"With Snowflake CoWork and Cortex Agents, leaders no longer have to rely on siloed portals or navigate disconnected tools to get answers. They can ask questions in natural language, directly from their browser or Microsoft Teams, and get trusted insights grounded in real enterprise data. What started as a digital transformation initiative has evolved into a fundamentally new way of working for Synopsys, and we're continuing to discover new use cases across the organization."

At Whoop, CoWork has broadened access to data across the company, Snowflake said.

"With Snowflake CoWork, we've reimagined how our teams interact with data. What used to require specialized analysts and manual requests is now accessible to hundreds of employees in real time," said Matt Luizzi, VP of Analytics at WHOOP.

"By automating routine queries and giving every team direct, trusted access to insights, we've reduced operational friction and enabled our data teams to focus on higher-impact work. The result is faster decisions, more efficient operations, and a scalable foundation for the next generation of AI at WHOOP."

Model training

Snowflake also used the announcement to expand its Cortex AI product family with Cortex Training. The service is designed to let companies customise open-weight foundation models, including models from the Qwen and Mistral families, using data already held within Snowflake.

The service gives customers managed GPU access for model training without requiring them to move sensitive data to external environments or manage distributed infrastructure themselves. Snowflake said early customer assessments show firms can complete up to twice as many training runs for the same GPU budget.

Resolve AI has made what Snowflake described as a multi-million-dollar commitment over two years to use Cortex Training for reinforcement learning on proprietary data. The company develops AI agents for operating software in production and is also a Snowflake supplier, with Snowflake engineering teams using Resolve AI in their own workflows.

"General-purpose API models will continue to improve, but there is a limit to how far you can get when building on top of them. Production operations for running and managing complex software systems have specific failure modes, reasoning patterns, and the highest bar for accuracy," said Spiros Xanthos, Founder and CEO of Resolve AI.

"We are overcoming these limitations with purpose-built training infrastructure, simulated large-scale environments, and evaluations based on real operational workflows. Cortex Training is a core part of how we scale these research priorities at Resolve AI."