TL;DR
Snowflake CoWork is Snowflake’s AI work agent for knowledge workers, rebranded from Snowflake Intelligence at Snowflake Summit 2026. It lets employees ask questions of governed enterprise data in plain language and get cited answers instead of raw dashboards. Beyond answering questions, CoWork automates multi-step work, builds shareable Artifacts from live data, and takes action across Gmail, Slack, Jira, and Salesforce through generally available MCP connectors. A context layer called Cortex Sense, still in private preview, grounds its answers in business definitions automatically. CoWork competes with Microsoft Copilot, Databricks Genie, and Salesforce Agentforce, but its accuracy depends entirely on how well a company’s Snowflake data is governed and modeled first.
It now reasons across enterprise data, cites its sources, and takes action in Gmail, Slack, Jira, and Salesforce without the user leaving the conversation. That shift matters because most employees never learned SQL, and most governed enterprise data still lives in systems only a handful of analysts know how to query.
This article breaks down what Snowflake CoWork actually does, how it evolved from Snowflake Intelligence, where it stands against Microsoft Copilot and Databricks Genie, and what a company needs in place before it delivers real value.
Key Takeaways Snowflake CoWork is the June 2026 rebrand of Snowflake Intelligence, expanded from a BI question-answering tool into a personal work agent that acts on enterprise data. CoWork reasons over governed Snowflake data, cites its sources, and takes action across Gmail, Slack, Jira, and Salesforce through generally available MCP connectors. Artifacts and shared conversations, live references to charts and tables that refresh under each viewer’s own data permissions, reached general availability on June 17, 2026. Cortex Sense, the layer that automatically grounds CoWork in business definitions, remains in private preview with no announced general availability date. CoWork’s accuracy depends on Snowflake data governance and semantic modeling being done well first, which is where most enterprise rollouts stall.
What Is Snowflake CoWork Snowflake CoWork is an AI agent built into the Snowflake platform that lets any employee ask questions of enterprise data in natural language and get a cited, governed answer back. It sits at ai.snowflake.com and reasons across structured data in Snowflake tables alongside unstructured sources like call transcripts, documents, and social mentions.
The agent does not stop at answering questions. It can draft a PowerPoint, update a Slack channel, or push a note into Salesforce based on what it finds, and it routes each request to the right tool automatically instead of asking the user to pick an agent.
A few things set CoWork apart from a generic chatbot layered on top of a warehouse.
Cited answers grounded in governed Snowflake data, not open web search results Persistent memory that learns a user’s preferences across repeated sessions Personal Skills that let a workflow be captured once and reused indefinitely Native iOS access with Face ID login, under the same governance as the desktop session
CoWork replaced Snowflake Intelligence in June 2026, and the rename reflects a real change in scope, not just a marketing refresh. The next section walks through what changed and why.
Kanerika’s Snowflake Consulting and Migration Services Kanerika is a Snowflake Select Tier Partner that builds the governed, well-modeled Snowflake environments that make tools like CoWork generate accurate code in the first place.
Explore Snowflake Services →
From Snowflake Intelligence to Snowflake CoWork 1. Snowflake Intelligence’s Launch and Early Growth Snowflake first unveiled Snowflake Intelligence at BUILD 2024, describing it as a platform that would let business users create data agents to analyze, summarize, and act on enterprise data. It reached general availability in November 2025, opening it to Snowflake’s full customer base at the time.
In its first three months at general availability, more than 1,000 customers deployed over 15,000 AI agents on the platform, according to Snowflake’s own announcement . That adoption curve is part of why Snowflake widened the product’s charter within seven months of GA.
2. The Summit 2026 Rebrand and What It Added Snowflake renamed the product CoWork at Summit 2026 on June 2, alongside Cortex Code becoming CoCo. Both names came from what internal teams were already calling the products informally, so Snowflake made the labels official rather than inventing new ones.
New capability shipped alongside the rename. This was not just a new name on the same product.
Artifacts, persistent chart and table references that refresh live and respect each viewer’s own permissions Cortex Sense, a context layer built from query history, object metadata, and BI dashboards MCP connectors to Gmail, Slack, Jira, and Salesforce Deliverable generation for PowerPoint, PDF, and Google Doc outputs directly from a conversation
Those additions are the difference between a tool that answers questions and one that finishes work. The features below cover each in more depth.
Key Features of Snowflake CoWork CoWork’s feature set breaks into five parts. Together they cover access, action, collaboration, context, and memory, each addressing a different point of friction in how employees have historically reached enterprise data.
1. Natural Language Access to Governed Data CoWork lets a non-technical employee type a business question and get an answer sourced directly from Snowflake tables, with citations attached. The agent respects existing role-based access controls, so a user only sees what their Snowflake permissions already allow. This removes the SQL bottleneck that keeps most enterprise data locked behind a small analytics team.
2. Action Across Business Tools Via MCP Connectors CoWork connects to Gmail, Slack, Jira, and Salesforce through Model Context Protocol connectors that reached general availability around the Summit announcements, according to Snowflake’s own recap . A user can ask CoWork to summarize an account in Salesforce, then have it draft a follow-up email in Gmail, without switching applications. Enterprises building custom AI applications on Snowflake often extend this same connector pattern to internal tools MCP does not cover out of the box.
3. Artifacts and Shared Conversations An Artifact is a live reference to a chart or table generated inside a CoWork conversation, and it stays connected to the source data. A teammate who opens a shared Artifact sees the same visualization filtered through their own permissions, not a static export. Artifacts and shared conversations both reached general availability on June 17, 2026 .
The pattern echoes how Kanerika’s own Karl analytics agent surfaces conversational insights inside client environments, though Karl runs independently of Snowflake’s agent stack.
4. Cortex Sense and the Context Layer Cortex Sense automatically builds a shared context layer from query history, table metadata, and dashboards in tools like Power BI and Tableau. Snowflake reports that Cortex Sense lifted CoCo and CoWork accuracy on complex enterprise queries from 47% to 83% in internal testing, compared with 23% for generic coding agents using Snowflake’s MCP server, per Snowflake’s product page . That figure is Snowflake’s own benchmark rather than an independently audited result, and Cortex Sense remains in private preview with no public general availability date.
Building an equivalent context layer outside a single vendor’s private preview typically falls under data strategy and semantic modeling work, not something teams can shortcut by waiting for a feature flag.
5. Personal Skills and Persistent Memory CoWork remembers a user’s preferences across sessions and lets them save a workflow once as a Skill, then reuse it without rebuilding it from scratch. Skills are moving toward public preview, according to Snowflake. That persistence is what separates a work agent from a one-off chatbot session that forgets everything between conversations.
Kanerika’s own DokGPT agent applies a similar memory pattern for document-based queries, independent of the Snowflake platform, which is a useful reference point for what persistent context looks like outside a single vendor’s ecosystem. CoWork does not exist in a vacuum. Before comparing it to other agents, it helps to know what it actually costs to run.
How Snowflake CoWork Compares to Other AI Work Agents Snowflake, Microsoft, Databricks, and Salesforce are all building the same basic idea from different starting points. Each is building a natural language interface over governed data, with a semantic layer underneath deciding how trustworthy the answer is.
1. Snowflake CoWork Vs Microsoft Copilot Microsoft Copilot lives inside the applications employees already open daily: Word, Excel, Outlook, and Teams. CoWork lives inside Snowflake and reaches out to those same tools through connectors instead of starting from them. For a company whose governed data already sits in Snowflake, that gives CoWork an advantage Copilot cannot fully replicate without a separate Microsoft Fabric integration effort, though business users who live in Microsoft 365 all day may find a second interface an added step.
2. Snowflake CoWork Vs Databricks Genie Both companies are chasing the same idea from different platforms: a natural language interface over governed data with a semantic layer underneath it. Genie ties into Databricks’ Unity Catalog and lakehouse architecture, while CoWork ties into Snowflake’s Horizon Context and Cortex Sense. The practical choice usually comes down to which platform already holds an enterprise’s governed data, not which agent has more features on paper.
3. Snowflake CoWork Vs Salesforce Agentforce Agentforce is built for CRM-centric workflows and runs on Salesforce’s own Data Cloud and Atlas reasoning engine. CoWork’s center of gravity is enterprise data broadly, not customer records specifically. A company already deep in Salesforce for sales and service workflows may lean on Agentforce for those tasks while using CoWork for company-wide data questions that touch more than the CRM.
Kanerika’s AI governance work spans engagements across all three platforms, since permission models rarely map cleanly from one vendor’s agent to another. The comparisons matter less than what CoWork is actually being used for inside real organizations. That’s next.
Agent Home Platform Primary Data Grounding Takes Action In Governance Model Snowflake CoWork Snowflake (ai.snowflake.com) Governed Snowflake data plus Cortex Sense context Gmail, Slack, Jira, Salesforce via MCP Snowflake role-based access controls Microsoft Copilot Microsoft 365 Microsoft Graph and Work IQ Word, Excel, Outlook, Teams Microsoft Entra ID and Purview Databricks Genie Databricks Lakehouse data plus Unity Catalog semantics Databricks Apps and dashboards Unity Catalog governance Salesforce Agentforce Salesforce Salesforce Data Cloud Sales Cloud, Service Cloud, Slack Salesforce profiles and permission sets
Real Enterprise Use Cases for Snowflake CoWork 1. Turning Unstructured Signals Into Business Action CoWork is built to reason over unstructured business context, not just structured tables. Call transcripts, support tickets, and social mentions of a brand can all become inputs the agent works from alongside Snowflake’s structured data, closing the gap between insight and action that Snowflake highlighted at Summit 2026 . Kanerika’s own context-aware AI agent work follows the same principle, grounding recommendations in operational context rather than a single structured source.
2. Cross-Sell and Audience Segmentation Fanatics uses Snowflake CoWork to help business users build accurate customer segments and addressable advertising audiences without waiting on a data team, and to surface cross-sell opportunities across the business, according to a statement in Snowflake’s own product materials . Snowflake also names Synopsys, Whoop, and Under Armour among early teams using CoWork to move faster from insight to action.
That segmentation use case is common in Kanerika’s own retail and FMCG engagements, where sales teams need audience data they can trust without routing every request through a central analytics queue.
3. Proactive Reporting Without Static Exports Because Artifacts refresh live and respect each viewer’s own data permissions, a team can share one Artifact across a group instead of exporting separate static reports that go stale within days. That single change removes the most common reason a number in a shared deck stops matching the number in the source system, a gap Kanerika sees constantly in finance and operations teams still waiting on a monthly close for numbers everyone else already has.
Every one of those use cases assumes the underlying data is already trustworthy. That assumption is where most CoWork rollouts either succeed quietly or stall in public.
What Enterprises Need Before Snowflake CoWork Delivers Value 1. Data Governance and Access Control CoWork inherits whatever access model already exists in Snowflake, so a poorly governed environment produces confidently wrong answers at agent speed instead of human speed. Role-based access controls, data classification, and clean object metadata all need to be in place before an agent sits on top of them. CIOs should stay cautious here, since poor governance makes an AI colleague confidently wrong at scale rather than obviously wrong, per TechFinitive’s Summit coverage .
Kanerika’s banking data governance work built on Microsoft Purview is a useful reference point for what that access model needs to look like before any agent gets near it, regulated industry or not. Companies unsure where their own governance stands can start with Kanerika’s AI Maturity Assessment before evaluating an agent rollout.
2. A Verified Semantic and Context Layer Cortex Sense promises to reduce the manual setup a semantic layer normally requires, but it draws its context from existing query history, metadata, and BI dashboards. If those inputs are inconsistent or undocumented, Cortex Sense inherits that inconsistency instead of correcting it.
What that foundation actually requires in practice.
Business definitions for core metrics, like revenue, fiscal calendars, and snapshot tables, documented and applied consistently BI dashboards in Power BI or Tableau connected, current, and free of conflicting metric definitions Object metadata in Snowflake tagged and maintained rather than left at platform defaults
This is not a CoWork-specific problem. It is the same governance foundation any enterprise AI agent depends on, whether it runs on Snowflake, Databricks, or Microsoft 365, and it is the same foundation Kanerika’s RAG development engagements build before layering a retrieval or agent interface on top.
Snowflake Readiness: How Kanerika Cut Manual Reconciliation by 60 Percent Kanerika is a Snowflake Select Tier Partner and an AI-first data engineering and automation consulting firm holding ISO 27001, ISO 27701, and SOC 2 Type II credentials, the kind of certifications enterprise data teams look for before granting an agent broad access to a Snowflake environment. The firm has built more than a decade of production data engineering and governance work, the layer that platforms like Snowflake CoWork depend on to deliver accurate answers rather than confident guesses.
Kanerika’s Snowflake practice covers migration, governance, and semantic modeling. Those are the same three layers CoWork and Cortex Sense both lean on.
Migrating and consolidating Snowflake environments through structured migration work so a single governed data layer exists instead of scattered, duplicated tables Building governance frameworks for role-based access control and data classification, extending Microsoft Purview practices to Snowflake environments Modeling business definitions and metrics consistently across BI tools so an agent layer like CoWork inherits clean context instead of conflicting numbers Applying AI and ML capabilities and predictive analytics on top of that governed layer once agent tools are actually ready to use it
FLIP replaces the manual schema mapping and validation work a Snowflake migration normally requires, the part of the project that typically eats the most engineering time and introduces the most human error.
Transform Your Snowflake Data Estate with AI-Ready Solutions! Partner with Kanerika for Expert Snowflake Migration and Governance Services
Book a Meeting
Case Study: Cost Savings With Snowflake Migration for Analytics Challenges Recurring annual subscription costs created financial inefficiencies with no long-term ownershipLimited refresh capability restricted reporting to hourly updates, slowing operational decision-making SSAS architecture constrained shareholder access and caused frequent outages, impacting reliability
Solutions Migrated the reporting environment from SSAS to Snowflake, eliminating licensing costs and enabling ownership Enabled table-level refreshes, delivering near real-time insights for critical operational reporting Integrated directly with shareholder ERP systems on Snowflake, expanding access from 43 shareholders to full shareholder coverage
Results 28% Annual Cost Savings 45% Faster Refresh Cycles 50% Fewer Outages
Wrapping Up Snowflake CoWork is Snowflake’s clearest bet yet that the next interface for enterprise data will be conversational, not a dashboard. The rebrand from Snowflake Intelligence added real capability, from Artifacts to MCP-connected actions to a context layer in Cortex Sense. None of it works well on top of ungoverned data.
A company evaluating CoWork should treat the agent as the last step, not the first, and get its Snowflake governance and semantic layer in order before rolling it out to knowledge workers who will trust whatever it tells them.
FAQs What is Snowflake CoWork? Snowflake CoWork is Snowflake’s AI work agent for knowledge workers, rebranded from Snowflake Intelligence at Snowflake Summit 2026. It lets users ask questions of governed enterprise data in natural language, get cited answers, and take action across tools like Gmail, Slack, Jira, and Salesforce, without writing SQL or leaving the conversation.
Is Snowflake CoWork the same as Snowflake Intelligence? Snowflake CoWork is the rebranded, expanded version of Snowflake Intelligence. Snowflake Intelligence reached general availability in November 2025 as a business intelligence question-answering tool, following an initial private preview announced back in late 2024. CoWork, introduced in June 2026, adds Artifacts, MCP-connected actions, Cortex Sense context, and persistent memory on top of that same foundation.
How does Snowflake CoWork access data securely? CoWork inherits a user’s existing Snowflake role-based access controls, so it only surfaces data that person’s account already has permission to see, rather than granting the agent its own broader access. Every answer cites its source in Snowflake, and administrators can disable sharing of Artifacts and chats for an entire account from CoWork’s settings inside Snowsight.
What is Cortex Sense and does it affect CoWork's accuracy? Cortex Sense is a context layer that automatically builds a shared understanding of business definitions from query history, metadata, and BI dashboards, without requiring manual semantic-model setup. Snowflake reports it lifts CoWork’s accuracy on complex queries from 47% to 83% in internal testing. It remains in private preview with no announced general availability date.
How is Snowflake CoWork different from Microsoft Copilot? Microsoft Copilot lives inside Microsoft 365 apps like Word, Excel, and Outlook, reasoning primarily over Microsoft Graph data and the work patterns captured in Work IQ. Snowflake CoWork lives inside Snowflake and reaches into external tools like Gmail and Salesforce through MCP connectors instead. The better fit depends on where an organization’s governed data already lives.
Which tools can Snowflake CoWork take action in? Snowflake CoWork connects to Gmail, Slack, Jira, and Salesforce through Model Context Protocol connectors that reached general availability around Summit 2026, with more connectors expected as the MCP ecosystem grows. Within those tools, CoWork can draft messages, update records, and generate deliverables like PowerPoint decks, PDFs, and Google Docs based on Snowflake data.
What does an enterprise need before deploying Snowflake CoWork? CoWork’s answers are only as reliable as the governance and semantic modeling underneath them. Enterprises need role-based access controls, tagged object metadata, documented business definitions, and connected, current BI dashboards in place first. Without that foundation, an agent can return confidently wrong answers faster than a human ever would, and at greater scale.
Is Snowflake CoWork generally available? Snowflake CoWork itself is available now at ai.snowflake.com to Snowflake customers with the agent enabled on their account. Some components sit at different release stages. Artifacts and shared conversations reached general availability on June 17, 2026, while Cortex Sense and Skills remain in private or public preview with no confirmed general availability date as of this writing.
