Snowflake recasts its AI strategy around action, not answers, with CoWork

Snowflake is adding workflow automation, multi-agent orchestration, and persistent user context to its AI-based enterprise data query platform, Intelligence — and renaming it CoWork.

It’s a sign the company wants to move beyond simply generating insights and help CIOs translate their AI investments into operational outcomes, analysts said.

Snowflake is previewing a new User Skills feature in CoWork with which developers will be able to turn routine tasks into automated workflows, then use CoWork’s MCP connectors (now generally available) to take those actions across multiple systems.

Independent consultant David Linthicum welcomed this ability to take actions without leaving CoWork, as it reduces handoffs between analytics, apps, and operations.

Another CoWork feature, multi-agent orchestration, is designed to break down complex requests and route work among different agents without requiring users to manually select or manage them, Snowflake said.

That could be one of the more strategically significant additions for CIOs, Linthicum said, as most enterprise workflows span retrieval, reasoning, approvals, actions, and follow-up.

“Snowflake is trying to package that complexity behind a single user experience. For CIOs, the appeal is better productivity and process continuity. The risk is that orchestration becomes difficult to audit, tune, and control at scale,” he said.

Dion Hinchcliffe, lead of the CIO practice at The Futurum Group, said it appears that Snowflake is moving from passive analytics and query assistance toward digital labor orchestration around enterprise data. “The strategic distinction is that copilots largely answered questions, while agentic systems are increasingly expected to execute workflows, coordinate tasks, and operate semi-autonomously across systems,” he said.

Giving AI agents business context

As part of that broader push toward agentic workflows, Snowflake plans soon to offer a private preview of Cortex Sense, a metadata and context layer designed to give AI agents a consistent understanding of enterprise data, business definitions, and operational knowledge.

This will automatically derive context from an enterprise’s data estate and make it available across agents and applications, reducing the need to recreate business context for individual workflows, the company said.

For CIOs, Cortex Sense could improve consistency, reduce the risk of hallucinations, and make AI outputs more operationally useful as shared context is what makes enterprise agents trustworthy, because it aligns business meaning, governance, and execution patterns across teams, Linthicum said.

Embedding business semantics, workflow intelligence, and agent skills all in the same vendor-specific orchestration layer could lead to high switching costs later for those wanting to change platforms, Hinchliffe warned: “Enterprises are likely entering an era where semantic lock-in may become as strategically important as data lock-in once was.”

Snowflake is not the only software vendor looking to create a semantic context layer to improve AI-driven business workflows. Cortex Sense has similarities in this respect with Microsoft’s recently released concept of ontologies, Salesforce’s metadata-driven AI layers, and ServiceNow’s workflow context models, Hinchcliffe said.

“All vendors are converging toward a shared realization: Enterprise AI systems require persistent organizational context to scale effectively,” Hinchcliffe said.

“What differs is where each vendor starts from strategically. Microsoft begins from productivity and platform dominance, ServiceNow from workflows, Salesforce from CRM and metadata, and Snowflake from governed enterprise data,” Hinchcliffe added.

Tougher platform choices ahead

That convergence will lead to a tussle between software vendors and hyperscalers to become the control plane for AI across enterprises.

“The long-term enterprise value in AI will likely accrue to vendors controlling orchestration, governance, policy enforcement, and workflow execution rather than raw model access alone,” Hinchcliffe said.

“Frontier models are increasingly commoditizing at the infrastructure layer. The durable enterprise differentiation is shifting toward context management, trust, operational integration, and orchestration,” he added.

The new features in CoWork don’t stop there: a new type of resusable artifacts, the conversational dashboard, will enable users to interact with governed views of live data, while Cortex Training, soon in public preview, will enable enterprises to train and customize foundation models using Snowflake-managed GPUs tailored to their own data and business requirements.

The conversational dashboards will move Snowflake into the overlap among analytics, copilots, and workflow automation, said Mike Leone, principal analyst at Moor Insights and Strategy.

“Power BI and Tableau are getting agentic, productivity copilots are extending into enterprise data, and CoWork is moving into productivity territory, all heading toward the same agent surface for knowledge workers,” he said.

The increasingly similar ambitions of analytics, productivity, and AI vendors will make it challenging to distinguish between competing platforms.

Linthicum suggested CIOs ask vendors whether their offerings just surface insights or can complete work autonomously.

However, if a company’s governed data and policy logic already live in Snowflake, CoWork may become the natural choice.

“Otherwise, incumbent analytics and productivity platforms still have serious distribution and workflow advantages,” Linthicum said.

CIOs looking for greater control over how AI is built and deployed may find value in Cortex Training, said Hinchcliffe.

“Many enterprises are becoming uncomfortable with complete dependence on external frontier model APIs for strategic workloads. Concerns around governance, sovereignty, cost predictability, latency, customization, and intellectual property are pushing enterprises toward more controllable AI architectures,” he said.