Snowflake CoWork, formerly Snowflake Intelligence, is Snowflake’s personal AI work agent for knowledge workers, announced June 2, 2026. According to Snowflake (2026), 13,600+ accounts now use Snowflake AI weekly. Business users query governed data, generate cited reports, and take action across Gmail, Slack, and Salesforce in natural language. Trustworthy answers depend on the quality of the context layer underneath.
Quick Facts
Permalink to “Quick Facts”| Field | Value |
|---|---|
| What It Is | Personal AI work agent for knowledge workers: query data, generate reports, automate tasks, take action |
| Formerly Known As | Snowflake Intelligence (fully rebranded June 2, 2026; existing customers migrated automatically) |
| Announced | June 2, 2026 - Snowflake Summit 26 |
| Access Point | ai.snowflake.com |
| Key Features | Deep Research, Artifacts, Cortex Sense, User Memory, Automations, User Skills, Skill Catalog, MCP Connectors |
| Availability | Partially GA; core chat GA; Artifacts, Deep Research, Cortex Sense, User Memory in preview |
| Integrates With | Gmail, Slack, Salesforce, Jira, Google Drive, Microsoft Excel, iOS |
| Reasoning Backbone | Anthropic (Claude) + OpenAI, Google, Meta, Mistral, DeepSeek, SpaceXAI |
| Atlan Relationship | CoWork launch partner; Atlan = 2025 Snowflake Data Governance Partner of the Year; Atlan MCP server feeds Enterprise Data Graph context to CoWork at inference time |
Inside Atlan AI Labs & The 5x Accuracy Factor
Learn how context engineering drove 5x AI accuracy in real customer systems. Explore real experiments, quantifiable results, and a repeatable playbook for closing the gap between AI demos and production-ready systems.
Download E-BookWhat is Snowflake CoWork? How the Context Layer Makes It Work in Production
Permalink to “What is Snowflake CoWork? How the Context Layer Makes It Work in Production”Snowflake CoWork is Snowflake’s personal AI work agent for knowledge workers: the rebrand of Snowflake Intelligence, announced June 2, 2026. According to Snowflake (2026), CoWork accounts more than doubled quarter-over-quarter, signaling a broader shift in how enterprises are adopting AI agent-powered data access. The product gives business users a natural-language interface to query governed Snowflake data, build reports, automate recurring tasks, and act across external tools like Gmail and Salesforce. The core chat interface is generally available at ai.snowflake.com.
The rebrand is complete: Snowflake Intelligence is now Snowflake CoWork. Existing customers are automatically migrated with no configuration changes and no migration window. What changed is not the underlying infrastructure but the strategic framing and product scope.
“CoWork” signals a deliberate shift in identity: from conversational analytics (“Intelligence”) to personal AI agency (“Work”). Snowflake has aligned CoWork with Anthropic’s Claude as the primary reasoning backbone, replacing the earlier Cortex-only architecture with a multi-model stack that includes OpenAI, Google, Meta, Mistral, DeepSeek, and SpaceXAI. The product now competes directly in the category that tools like Claude for Work and Microsoft Copilot are defining. According to CIO.com (2026), the industry shift is “from copilots that largely answered questions to agentic systems executing workflows autonomously,” in the words of Dion Hinchcliffe, Futurum Group.
CoWork operates as a consumption layer. Knowledge workers (analysts, sales operations, HR, finance) interact with enterprise data without writing [SQL or building pipelines. The semantic layer The interface is available on web today, with an iOS app, Microsoft Excel extension, and Slack bot coming to preview and GA shortly. CoWork also supports Domain Agents that auto-route questions to the most relevant data agent based on business domain.
CoWork’s launch comes against a backdrop of unmet enterprise AI expectations. According to SP Global (2026), 49% of organizations say generative AI has not delivered on its promises — the context gap is a key driver of AI agent hallucination. CoWork is Snowflake’s answer to that gap, moving from text responses to governed, actionable, and cited enterprise intelligence. To understand what CoWork can do in practice, it helps to examine its core capabilities one by one.
What are Snowflake CoWork’s Key Features?
Permalink to “What are Snowflake CoWork’s Key Features?”Snowflake CoWork ships 15+ named capabilities at Summit, organized into five functional clusters: research and reasoning, publishing, personalization, automation, and integration. The core chat interface and Domain Agents are generally available; most new features (Deep Research, Artifacts, Cortex Sense, User Memory, and Automations) are in preview as of June 2026. According to the Snowflake blog (2026), Skill Catalog and Agent Studio are also coming to GA soon.
| Feature | What It Does | Current Availability |
|---|---|---|
| Core Chat Interface | Natural-language Q&A over Snowflake-governed structured and unstructured data | Generally available |
| Deep Research | Multi-step reasoning with fully cited, downloadable reports | GA soon |
| Artifacts | Saved, shareable interactive dashboards backed by live governed data | GA soon |
| Cortex Sense | Runtime context enrichment - auto-unifies data, business definitions, and activity signals for agents | Private preview soon |
| User Memory | Personalizes responses based on user behavior and query history | Public preview soon |
| User Skills | Reusable, automated multi-step workflows users can build and share | Public preview soon |
| Skill Catalog | Team-wide discovery and reuse of skills across the organization | GA soon |
| Automations | Scheduled briefs, anomaly alerts, recurring analysis tasks | Public preview soon |
| Agent Studio | Build, evaluate, and monitor agents with cost controls | GA soon |
| Agent Router | Auto-routes questions to the most relevant domain agent | Private preview soon |
| Domain Agents | Auto-directs requests to the most relevant data agent by domain | Generally available |
| MCP Connectors | Integrations with Slack, Google Drive, Salesforce via open MCP standard | Generally available |
| File Output | Generates PowerPoint, PDF, and Google Docs from analysis | Public preview soon |
| iOS Mobile App | Face ID login, mobile access to CoWork | GA soon |
| Slack Bot | Morning data briefings delivered directly in Slack | Private preview soon |
| Microsoft Excel Extension | Data analysis inside Excel | Private preview soon |
Artifacts
Permalink to “Artifacts”Artifacts are publishable, shareable analyses: governed interactive dashboards that live and refresh from Snowflake data. Business users can build, share, and embed them without BI tool overhead. Artifacts solve the “last-mile” problem of AI output: instead of a text answer that expires in the chat window, the insight becomes a durable, governed artifact other teams can consume. Status: GA soon.
Cortex Sense
Permalink to “Cortex Sense”Cortex Sense is CoWork’s runtime context enrichment layer. It automatically builds signals from data and activity already in Snowflake (usage patterns, user role, historical feedback) and injects that context into agent responses. Cortex Sense is the mechanism Snowflake intends to use to close the accuracy gap in CoWork answers. However, Cortex Sense is in private preview only as of June 2026. CoWork ships to enterprises before this capability is generally available, meaning every CoWork deployment at Summit GA operates without Snowflake’s own context infrastructure at production readiness.
Deep Research
Permalink to “Deep Research”Deep Research is CoWork’s multi-step reasoning mode. CoWork breaks down a complex question, pulls from multiple data sources, synthesizes findings, and produces a fully cited report. This removes the need for a senior analyst to stitch together multi-source reports manually. Status: GA soon.
User Memory and Automations
Permalink to “User Memory and Automations”User Memory learns from how a user interacts with data over time: preferred metrics, reporting cadence, tolerance for detail. Responses become more relevant without additional prompting. Automations run scheduled briefs and anomaly alerts without user initiation. Both are in public preview.
MCP integrations: Gmail, Slack, Salesforce, iOS
Permalink to “MCP integrations: Gmail, Slack, Salesforce, iOS”CoWork connects to external tools via what is MCP (Model Context Protocol), an open standard, not a proprietary integration. GA as of Summit: Slack, Google Drive, and Salesforce. Gmail, Jira, Microsoft Excel extension, and the iOS app are coming to preview and GA shortly. Snowflake has announced intent to acquire Natoma to add identity-aware governance to MCP connections, addressing the security gap that currently limits enterprise MCP adoption at scale.
How does Snowflake CoWork Work?
Permalink to “How does Snowflake CoWork Work?”Snowflake CoWork routes a user’s natural-language question through four steps: Snowflake Cortex retrieves relevant structured and unstructured data, Cortex Analyst generates the SQL or reasoning plan, an Agent Router directs the query to the right domain agent, and Anthropic’s Claude model (with available context) produces a cited response. Snowflake Horizon governance policies (RBAC, row-level security, and data masking) apply at every layer.
A single user question moves through CoWork in sequence:
- The user asks in natural language via web, iOS, Slack, or Excel
- Agent Router identifies the relevant domain agent (Sales, Finance, Operations)
- Cortex Search pulls relevant documents and structured data from Snowflake
- Cortex Analyst generates SQL for questions requiring structured data
- Cortex Sense (when available) enriches context at runtime with user-role signals and activity data
- Anthropic Claude reasons over the retrieved context and generates a cited response
- The response returns as chat, is saved as an Artifact, or triggers an Automation
CoWork differs from generic AI assistants in two important ways. First, data access is governed: existing Snowflake RBAC, row-level security, and data masking policies apply. Users cannot access data they do not have permission for. Second, CoWork queries live data rather than a static training corpus, unlike RAG systems that index data offline, so answers reflect the current state of enterprise data and cite their sources.
There is a gap CoWork does not solve natively. CoWork’s governance is Snowflake-scoped. Cross-system context (dbt metrics, Tableau definitions, Salesforce field semantics) is not natively available to CoWork. Cortex Sense handles Snowflake-internal signals; it does not bridge definitional inconsistencies across systems, such as “revenue” defined differently in Salesforce versus Snowflake versus the data warehouse. Cross-system lineage requires a context layer for Snowflake that reaches beyond the warehouse boundary.
Snowflake CoWork vs. Snowflake Intelligence, Microsoft Copilot, and Databricks
Permalink to “Snowflake CoWork vs. Snowflake Intelligence, Microsoft Copilot, and Databricks”Snowflake CoWork replaces Snowflake Intelligence with a wider product scope: adding Artifacts, User Memory, multi-model support, Anthropic Claude as the reasoning backbone, and MCP integrations. Against Microsoft Copilot, CoWork’s moat is governed enterprise data access; Copilot’s strength is workplace ubiquity. Against Databricks, both platforms share the same context gap: neither resolves cross-system business logic without a dedicated context layer, as analyzed by SiliconANGLE (2026).
CoWork competes in three dimensions: with its own predecessor (Snowflake Intelligence), with Microsoft’s productivity-layer dominance (Copilot), and with Databricks’ data-platform equivalent (Genie/AI BI).
CoWork vs. Snowflake Intelligence: what changed in the rebrand
Permalink to “CoWork vs. Snowflake Intelligence: what changed in the rebrand”| Dimension | Snowflake Intelligence | Snowflake CoWork |
|---|---|---|
| Core identity | Conversational analytics interface | Personal AI work agent |
| Reasoning backbone | Cortex (Snowflake’s own models) | Anthropic Claude + multi-model (OpenAI, Google, Meta, Mistral) |
| Automation | Limited | Automations, User Skills, scheduled briefs |
| Publishing | Chat output only | Artifacts: shareable governed dashboards |
| Memory | None | User Memory (personalizes over time) |
| External integrations | None | MCP Connectors: Slack, Google Drive, Salesforce, Gmail (coming), Jira (coming) |
| Mobile | None | iOS app with Face ID (GA soon) |
| Existing customers | N/A | Automatic migration, no action required |
CoWork vs. Microsoft Copilot
Permalink to “CoWork vs. Microsoft Copilot”| Dimension | Snowflake CoWork | Microsoft Copilot (365) |
|---|---|---|
| Primary data source | Governed Snowflake data + MCP connectors | Microsoft 365 Graph (email, calendar, Teams, SharePoint) |
| Governance model | Snowflake RBAC, row-level security, masking | Microsoft Purview integration |
| User interface | ai.snowflake.com + Slack + iOS | Native in Teams, Outlook, Word, Excel |
| Ubiquity | Requires Snowflake ecosystem adoption | Already deployed wherever Microsoft 365 is |
| Structured data access | Deep: all Snowflake-governed tables | Surface-level; depends on Microsoft Fabric integration |
| Cross-system context | Snowflake-native; gaps for dbt/Salesforce/BI | Microsoft Graph: productivity-scoped |
| Adoption barrier | Business users must log into Snowflake interface | Zero: users are already in Microsoft 365 |
| Moat | Governed enterprise data accuracy | Workplace integration depth and ubiquity |
CoWork’s adoption challenge is real. For organizations whose entire data workflow lives in Snowflake, CoWork provides governed, cited access that Copilot cannot replicate without extensive Microsoft Fabric integration work. But as CIO.com (2026) noted, business users who live in Microsoft 365 all day face genuine friction logging into a separate Snowflake interface, even when the data quality on the other side is better.
On Databricks: both Snowflake and Databricks are making the same strategic move (natural language interfaces over governed data, context catalog layers, and “system of intelligence” positioning). According to SiliconANGLE’s analysis (2026), both platforms share identical strategic gaps: neither has cross-system business logic modeling, real-time operational state, or institutional memory natively. Atlan serves both platforms.
How does Atlan’s Context Layer Make Snowflake CoWork Answers Trustworthy?
Permalink to “How does Atlan’s Context Layer Make Snowflake CoWork Answers Trustworthy?”Without a context layer, CoWork accuracy measures 47%, according to Atlan AI Labs (2026) internal measurement on Cortex queries. With Atlan’s Enterprise Data Graph feeding CoWork via MCP, accuracy reaches production-grade, representing a 5x improvement according to Atlan AI Labs (2026). Context Engineering Studio validates the semantic definitions CoWork sees before they ship. Context Agents auto-generate the business glossary and certified metrics CoWork depends on. Atlan is CoWork’s official launch partner and Snowflake’s 2025 Data Governance Partner of the Year, bringing AI agent governance to every CoWork deployment.
The accuracy gap CoWork ships with
Permalink to “The accuracy gap CoWork ships with”Snowflake CoWork answer quality depends entirely on the quality and completeness of context available at inference time. Cortex Sense (the native runtime enrichment layer) is in private preview only as of June 2026. Snowflake Horizon Context’s Business Glossary is H2 2026 roadmap. This means every CoWork deployment at Summit GA ships without the context infrastructure Snowflake’s own product requires to be trustworthy.
The accuracy data: Atlan’s internal measurement found 83% accuracy on Cortex queries with Cortex Sense context enrichment versus 47% without. The gap is context. Separately, Atlan’s BirdBench study (145 queries, p < 2e-10) demonstrated a 3x improvement in SQL accuracy on complex queries when Atlan metadata (column descriptions, glossary terms, cross-system lineage) was present versus a schema-only baseline.
What Atlan’s context layer provides
Permalink to “What Atlan’s context layer provides”Understanding context engineering for AI agents and how businesses structure and deliver semantic knowledge to AI models is essential to closing CoWork’s accuracy gap. The 5x accuracy improvement Atlan AI Labs measured comes directly from this discipline applied at inference time.
| What CoWork Needs | Snowflake Native (Cortex Sense) | Atlan Enterprise Data Graph |
|---|---|---|
| Consistent semantic definitions (“revenue” = same across Finance, GTM, Product) | Horizon Context Business Glossary: H2 2026 roadmap | Business Glossary: shipped; Active Ontology resolves conflicting definitions across tools |
| Cross-system lineage (dbt to Snowflake to Tableau to Salesforce) | Snowflake-internal lineage only | Column-level lineage across 100+ connectors |
| Runtime context delivery to agents | Cortex Sense (private preview, Snowflake signals only) | MCP server for Snowflake: delivers certified context to CoWork at inference time, GA |
| Context validation before production | Semantic Studio (private preview) | Context Engineering Studio: bootstrap, test, version, and CI-integrate context as code, GA |
| Auto-generated business descriptions | Cortex Sense runtime enrichment | Context Agents: 690K+ descriptions generated, 87% rated on par or better than human writing |
| Number of metadata connectors | 5 (Horizon Context Wave 1, private preview) | 100+ connectors |
Atlan’s four-product context flow for CoWork
Permalink to “Atlan’s four-product context flow for CoWork”Atlan is the context layer for Snowflake: the context plane that layers on top of Snowflake’s data plane. What constitutes a complete enterprise context layer is defined by cross-system coverage, certified semantics, and agent delivery. It is not a replacement for CoWork. It makes CoWork trustworthy. The broader architecture of what an enterprise context layer does across every AI system in a data stack is what Atlan is built for. Four products deliver this:
-
Enterprise Data Graph ingests metadata from every system CoWork touches: not just Snowflake, but dbt models, Fivetran pipelines, Tableau, and Salesforce. 100+ connectors, column-level lineage, certified definitions unified via the semantic layer across every tool.
-
Context Engineering Studio validates the semantic layer before it reaches CoWork. Teams bootstrap context as code, run CI-integrated eval suites, and version context like software. One insurance customer compressed a 1-year context build to 1 month using this approach.
-
Context Agents auto-generate the business glossary, ontology, and certified metrics CoWork depends on. 690K+ descriptions generated, 87% quality rating at or above human writing, across 50+ enterprise deployments.
-
Context Lakehouse (Atlan’s Iceberg-native context store) serves all of this to CoWork via MCP at inference time, from a single cross-system source of truth. The agent context layer architecture explains how this delivery works.
Context flows to every CoWork agent via MCP, the same open standard CoWork uses for Gmail and Salesforce integrations. Atlan is also a launch partner for Snowflake’s Open Semantic Interchange (OSI), delivering AI agent governance at the semantic level, the standard by which Snowflake and Atlan exchange semantic definitions bidirectionally.
Atlan’s Enterprise Data Graph is in production at 400+ enterprises representing $10T+ in market cap. The framing: Snowflake = data plane. Atlan = context plane.
Context for AI Analysts: See Atlan's Context Studio in Action
Context is what gets AI analysts to production. See how teams are building production-ready AI analysts with Atlan's Context Studio.
Save your SpotReal stories from real customers: enterprises making AI work in production
Permalink to “Real stories from real customers: enterprises making AI work in production”"We're excited to build the future of AI governance with Atlan. All of the work that we did to get to a shared language at Workday can be leveraged by AI via Atlan's MCP server...as part of Atlan's AI Labs, we're co-building the semantic layer that AI needs with new constructs, like context products."
- Joe DosSantos, VP of Enterprise Data & Analytics, Workday
"Atlan is much more than a catalog of catalogs. It's more of a context operating system...Atlan enabled us to easily activate metadata for everything from discovery in the marketplace to AI governance to data quality to an MCP server delivering context to AI models."
- Sridher Arumugham, Chief Data & Analytics Officer, DigiKey
CoWork is well-designed. Context is what makes it production-ready.
Permalink to “CoWork is well-designed. Context is what makes it production-ready.”Snowflake CoWork is a genuine step forward: governed AI access for knowledge workers, built on Anthropic Claude, with MCP integrations that reach across the enterprise toolchain. It is the right product for the right moment. The gap is not in CoWork’s design. It is in the context infrastructure that CoWork depends on to produce trustworthy answers, which Snowflake’s own stack does not yet provide at GA.
That gap is precisely where Atlan operates. The Enterprise Data Graph, Context Agents, Context Engineering Studio, and Context Lakehouse deliver the semantic foundation CoWork needs today, before Cortex Sense reaches GA and before Horizon Context’s Business Glossary ships in H2 2026. Teams implementing a context layer for AI agents today (not waiting for Snowflake’s native stack to mature) can follow the enterprise context layer implementation guide to get started.
The window between CoWork’s GA launch and Snowflake’s context infrastructure reaching production maturity is where enterprises are making decisions about what their AI agents are allowed to know, and how they know it.
FAQs about Snowflake CoWork
Permalink to “FAQs about Snowflake CoWork”1. Is Snowflake CoWork the same as Snowflake Intelligence?
Yes. Snowflake CoWork is the new name for Snowflake Intelligence, rebranded at Snowflake Summit on June 2, 2026. Existing customers are automatically migrated with no action required. The product has expanded from a conversational analytics interface to a full personal work agent, adding Artifacts, User Memory, Automations, multi-model support (Anthropic Claude as primary backbone), and MCP integrations with Gmail, Slack, and Salesforce.
2. What is the difference between Snowflake CoWork and Snowflake CoCo?
Snowflake CoCo (Cortex Copilot) is an AI coding and data engineering assistant embedded in Snowflake Notebooks and the SQL editor, designed for data engineers and analysts who write code. Snowflake CoWork is a natural-language work agent for knowledge workers (business analysts, operations, sales, and finance) who need governed data access without writing SQL. Both run on Cortex and Anthropic Claude; they address different user personas and different interaction patterns.
3. What can Snowflake CoWork do?
Snowflake CoWork can: answer natural-language questions over governed Snowflake data; run multi-step Deep Research and generate fully cited reports; create and publish Artifacts (interactive governed dashboards); automate recurring briefs and anomaly alerts; connect to Gmail, Slack, Salesforce, and Google Drive via MCP; and personalize responses over time using User Memory. Most features beyond the core chat interface are in preview as of June 2026.
4. How does Snowflake CoWork use Cortex Sense?
Cortex Sense is the runtime enrichment layer that automatically builds signals from data activity and user behavior already in Snowflake and injects that context into CoWork responses. Cortex Sense is in private preview as of the June 2026 announcement and is not yet generally available. Until Cortex Sense reaches GA, Atlan’s Enterprise Data Graph provides the semantic foundation CoWork needs for production-grade accuracy. The full picture of what a context layer for Snowflake Cortex must provide is detailed separately.
5. What is Deep Research in Snowflake CoWork?
Deep Research is CoWork’s multi-step reasoning mode. Instead of returning a single query result, CoWork decomposes a complex question, pulls relevant data from multiple sources, synthesizes findings, and delivers a fully cited report including charts, tables, and downloadable documents. It is the equivalent of assigning a senior analyst to compile a cross-functional research brief, but in real time. Deep Research is in GA-soon status as of June 2026.
6. Can Snowflake CoWork connect to Gmail and Salesforce?
Yes. CoWork uses the Model Context Protocol (MCP), an open standard, to connect to external tools. MCP Connectors for Slack, Google Drive, and Salesforce are generally available as of Summit. Gmail and Jira connectors are announced. Snowflake’s pending acquisition of Natoma will add identity-aware governance to these MCP connections, addressing enterprise security requirements around agent-to-app authorization.
7. Is Snowflake CoWork available now?
The core CoWork chat interface and Domain Agents are generally available as of June 2, 2026. MCP Connectors for Slack, Google Drive, and Salesforce are also GA. Deep Research, Artifacts, iOS Mobile App, Skill Catalog, and Agent Studio are GA soon. User Memory, Automations, User Skills, and File Output are in public preview. Cortex Sense, Agent Router, Slack Bot, and Microsoft Excel Extension are in private preview. Access is at ai.snowflake.com.
8. Why are Snowflake CoWork answers sometimes inaccurate?
CoWork answer quality depends on the quality of the context layer underneath it. Without consistent semantic definitions (business glossary), cross-system lineage, and certified metrics, CoWork can produce conflicting metric calculations and answers that do not reflect current data — including AI agent hallucination scenarios where confidently wrong answers reach business users. According to Atlan AI Labs (2026), internal measurement found 47% accuracy on Cortex queries without a context layer versus 83% with Cortex Sense context enrichment. Snowflake’s own context infrastructure (Cortex Sense and the Horizon Context Business Glossary) is in private preview or H2 2026 roadmap status. Atlan’s Enterprise Data Graph provides this foundation today.
Sources
Permalink to “Sources”- Snowflake CoWork Powers the Agentic Enterprise
- Snowflake CoWork: Personal Work Agent
- Snowflake recasts its AI strategy around action, not answers, with CoWork
- Snowflake moves up the AI stack – but the System of Intelligence is still being built
- Inside Atlan AI Labs: The 5x Accuracy Factor
- Making Talk to Data 3x More Accurate: How Snowflake Intelligence and Atlan
- Enterprise AI adoption research — SP Global (2026)
- Snowflake Intelligence 101: A Complete Overview
