Snowflake Cortex Sense is a runtime context enrichment layer announced June 2, 2026 at Snowflake Summit. It lifts CoCo and CoWork accuracy from 47% to 83% by automatically building a shared context substrate from query history, object metadata, BI dashboards, and Horizon Context semantic views. No manual configuration required: business definitions become available to agents at query time.
What is Snowflake Cortex Sense? How it connects to the enterprise context layer
Permalink to “What is Snowflake Cortex Sense? How it connects to the enterprise context layer”The core challenge Cortex Sense solves: enterprise AI agents know the data schema, but not the business meaning behind it. A query for “Q3 revenue” fails if the agent does not know the company’s fiscal calendar starts in February. Cortex Sense closes that gap at runtime, automatically, by enriching agents with signals already present in Snowflake.
Announced June 2, 2026 at Snowflake Summit, Cortex Sense builds a shared context layer substrate from query history, metadata, BI dashboard definitions, and Horizon Context semantic views, then delivers it to CoWork and CoCo the moment they need it.
Snowflake CPO Christian Kleinerman described it at the Summit keynote as a capability that “helps collect signals, enrich those signals and make them available to CoCo, CoWork or Cortex agents for you to get more context and more semantic information.”
Cortex Sense and its relationship to CoWork and CoCo
Permalink to “Cortex Sense and its relationship to CoWork and CoCo”Cortex Sense is the shared context substrate beneath two of Snowflake’s primary agent surfaces. CoWork is the personal work agent for knowledge workers (formerly Snowflake Intelligence), and Snowflake CoCo is the coding agent, formerly Cortex Code. Both were announced June 2, 2026 as part of the same Summit release. Both draw from the same Cortex Sense layer at runtime. Cortex Sense is not a standalone product: the agents are the interface, and Cortex Sense is the context engine underneath.
Cortex Sense and its relationship to Horizon Context
Permalink to “Cortex Sense and its relationship to Horizon Context”Horizon Context (part of Horizon Catalog) is the governance and metadata substrate: it manages semantic views, metadata connectors, and the business glossary layer. Cortex Sense is the runtime enrichment layer that activates those signals for agents at query time.
The distinction matters. Horizon Context is the library. Cortex Sense is the librarian that retrieves the right context for each agent at the moment of query. According to the Snowflake Horizon Context Blog (June 2026), Cortex Sense draws from whatever semantic definitions Snowflake Horizon Context currently holds, including a business glossary capability that Snowflake has on its H2 2026 roadmap.
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 SpotHow does Snowflake Cortex Sense work?
Permalink to “How does Snowflake Cortex Sense work?”Cortex Sense works by building a managed runtime context layer from four signal types already present in Snowflake, then serving that context to agents the moment they need it. No manual enrichment pipeline required.
Signal collection: what Cortex Sense draws from
Permalink to “Signal collection: what Cortex Sense draws from”According to the Snowflake CoWork Blog (June 2026), Cortex Sense builds its context substrate from four sources:
- Query history: Patterns (similar to RAG query signals) in how analysts query data reveal business definitions implicitly. If data scientists consistently filter by a particular date field, Cortex Sense infers that field as a key temporal anchor.
- Object metadata: Table names, column names, descriptions, and relationships registered in Snowflake’s schema.
- BI dashboard definitions: Power BI and Tableau dashboard metrics and measures: the closest proxy for business-validated semantic definitions already in Snowflake’s orbit.
- Horizon Context semantic views: The formal semantic definitions Snowflake’s governance layer manages: revenue definitions, fiscal calendars, snapshot table logic.
Context enrichment: what Cortex Sense builds
Permalink to “Context enrichment: what Cortex Sense builds”Cortex Sense unifies the four signal types into a single shared context layer. It understands complex constructs: revenue definitions (which finance and sales often define differently), fiscal calendars (when Q1 starts varies by company), and snapshot tables (point-in-time logic that breaks standard query patterns).
Enrichment is role-aware. As Snowflake noted in Summit field notes captured by the Atlan team, Cortex Sense contextualizes answers differently depending on who is asking: a data scientist and a sales analyst will receive the same data framed in the way their role typically uses it.
Agent delivery: how context reaches CoWork and CoCo
Permalink to “Agent delivery: how context reaches CoWork and CoCo”Both CoWork and CoCo draw from the same Cortex Sense layer at runtime. The shared substrate eliminates context drift between the two agent types: both answer from the same enriched foundation. For multi-agent architectures that extend beyond Snowflake, an agent context layer provides the cross-system equivalent of this shared enrichment.
Cortex Sense also activates Snowflake’s existing MCP server for Snowflake. According to the Snowflake CoWork Blog (June 2026), frontier coding agents connecting via Snowflake MCP alone achieve 23% accuracy on complex enterprise queries. With Cortex Sense, the same agents reach 83%.
Prebuilt role plugins for finance and sales bundle skills, business logic, and MCP connectors, enabling your team to deploy production-ready agents for those roles without rebuilding the context layer from scratch. Additional role plugins are on the roadmap.
| Agent behavior | Without Cortex Sense | With Cortex Sense |
|---|---|---|
| Revenue definition | Uses raw column name (“rev_usd_arr”) | Resolves to governed definition (“Annual Recurring Revenue, as defined in fiscal glossary”) |
| Fiscal calendar | Defaults to calendar year | Applies company-specific fiscal calendar (e.g., February start) |
| Role-specific framing | Same answer to data scientist and sales analyst | Contextualizes answer per user role |
| BI metric alignment | Ignores dashboard definitions | Ingests Power BI and Tableau measure definitions |
| Snapshot table logic | Returns latest row (incorrect) | Recognizes snapshot pattern, returns point-in-time row |
| Accuracy (complex queries) | 47% (CoCo/CoWork without Cortex Sense) | 83% (CoCo/CoWork with Cortex Sense) |
What does Snowflake Cortex Sense actually improve?
Permalink to “What does Snowflake Cortex Sense actually improve?”Snowflake’s internal benchmarks show a 36-percentage-point accuracy lift when Cortex Sense enriches agents: from 47% without it to 83% with it, measured on complex enterprise queries against production data environments.
The accuracy benchmark
Permalink to “The accuracy benchmark”According to the Snowflake CoWork Blog (June 2026), the improvement spans three reference points. A fourth data point from SiliconANGLE (June 2026) suggests a slightly different benchmark methodology but confirms the same directional improvement.
| Scenario | Accuracy | Source |
|---|---|---|
| Frontier coding agent + Snowflake MCP only | 23% | Snowflake CoWork Blog (June 2026) |
| CoCo/CoWork without Cortex Sense | 47% | Snowflake CoWork Blog (June 2026) |
| CoCo/CoWork with Cortex Sense | 83% | Snowflake CoWork Blog (June 2026) |
| Cortex Sense (independent benchmark) | 86% | SiliconANGLE (June 2026) |
The SiliconANGLE variance (86% vs. 83%) likely reflects a different evaluation set. Both figures represent a material improvement in agent trustworthiness for complex enterprise queries.
What “accuracy” means in this context
Permalink to “What “accuracy” means in this context”These benchmarks measure performance on complex enterprise queries, not toy data or simple lookups. Complex queries require understanding of business definitions, fiscal logic, and multi-table relationships. This is exactly the category where context enrichment has the greatest impact.
Baris Gultekin, VP of AI at Snowflake, stated at Summit: “Context determines agent quality.” The benchmark is the quantification of that principle.
According to SiliconANGLE (June 2026), Cortex Sense is “strongest today around understanding structured data.” The benchmark aligns with the capability’s design center.
Where accuracy improvement matters most
Permalink to “Where accuracy improvement matters most”- Finance use cases: Questions involving fiscal calendars, revenue definitions, and period comparisons are exactly the constructs Cortex Sense is designed to resolve.
- Sales use cases: Pipeline queries, account definitions, ARR vs. bookings distinctions. The prebuilt sales plugin targets this directly.
- Analytics queries: Snapshot table logic, cohort definitions, and metric roll-ups that require knowing the business rules, not just the schema.
Analyst David Linthicum (CIO.com, June 2026) noted that Cortex Sense “could improve consistency, reduce the risk of AI agent hallucination, and make AI outputs more operationally useful.” The mechanism is shared context that aligns business meaning across teams.
Snowflake Cortex Sense vs a cross-system context layer
Permalink to “Snowflake Cortex Sense vs a cross-system context layer”Cortex Sense enriches Snowflake agents with high accuracy, but only within Snowflake’s perimeter. Most enterprise data estates span 15 to 30 systems, and the context those agents need lives outside the warehouse.
What Cortex Sense covers (the honest picture)
Permalink to “What Cortex Sense covers (the honest picture)”Cortex Sense does real work within its scope. According to the Snowflake CoWork Press Release (June 2026) and Horizon Context Blog (June 2026), its coverage includes:
- Query history, metadata, and semantic views from within Snowflake’s data estate
- BI dashboards from Power BI and Tableau via Horizon Context connectors
- 5 metadata connectors in Wave 1 (private preview), a roadmap item Snowflake has not disputed
- Prebuilt role plugins for finance and sales
The “enrichment built on sand” problem
Permalink to “The “enrichment built on sand” problem”There is a risk that current SERP coverage does not address: Cortex Sense learns from and scales whatever context already exists in Snowflake. If revenue is defined inconsistently across finance and sales, or if the business glossary is sparse, Cortex Sense automates and accelerates the inconsistency.
This is not a defect in Cortex Sense’s design. It is the inherent constraint of runtime enrichment from a single-system foundation. According to SiliconANGLE (June 2026), Cortex Sense is “much weaker on capturing deep process knowledge” beyond structured query signals.
Analyst Dion Hinchcliffe, writing in CIO.com (June 2026), warned of “semantic lock-in”: embedding business semantics within Snowflake’s orchestration layer creates switching costs. A cross-system context layer keeps semantic definitions vendor-neutral and prevents that lock-in.
Cortex Sense vs a cross-system context layer: the comparison
Permalink to “Cortex Sense vs a cross-system context layer: the comparison”The right frame is additive, not adversarial. Cortex Sense is Snowflake’s runtime enrichment layer. A cross-system context layer for Snowflake Cortex is the foundation that makes Cortex Sense accurate across the full enterprise stack, not just within the warehouse. The agent context layer architecture enables this cross-system enrichment.
| Dimension | Cortex Sense | Atlan Enterprise Data Graph |
|---|---|---|
| Scope | Snowflake data estate + 5 metadata connectors (Wave 1) | 100+ connectors: warehouses, BI tools, pipelines, cloud services |
| Context source | Query history, object metadata, BI dashboards, semantic views | Column-level lineage, certified business glossary, ontology, quality signals |
| Business glossary | Via Horizon Context (H2 2026 roadmap) | Shipped: governed definitions across all connected systems |
| Lineage | Object-level within Snowflake | Column-level, cross-system (reverse-engineered from SQL) |
| Agent delivery | Runtime enrichment at query time for CoWork and CoCo | MCP server, A2A protocol, SQL interface, open APIs (any agent) |
| Context generation | Automated from existing Snowflake signals | Context Agents: 690K+ descriptions, 87% human quality, CI-validated |
| Status | Private preview (announced June 2, 2026) | GA across 50+ enterprise customers |
How Atlan’s Enterprise Data Graph works with Snowflake Cortex Sense
Permalink to “How Atlan’s Enterprise Data Graph works with Snowflake Cortex Sense”Atlan is the context layer for AI: the governed infrastructure that delivers enterprise knowledge to every model, every agent, and every team from a single source of truth. It layers directly on top of Snowflake, not alongside it.
Why Cortex Sense alone may not be enough for your full stack
Permalink to “Why Cortex Sense alone may not be enough for your full stack”According to Atlan AI Labs research, 83% of AI pilots never reach production. The gap is context, not the model.
Most enterprise data estates span 15 to 30 systems. A Snowflake agent asking about “customer churn” needs context from Salesforce (CRM definitions), dbt (transformation logic), and Looker (metric definitions), not just Snowflake schemas. Cortex Sense’s enrichment, however accurate within the warehouse, cannot close that cross-system gap.
Atlan is Snowflake’s 2025 Data Governance Partner of the Year and was a launch partner for Snowflake CoWork (then Snowflake Intelligence). The relationship is established and the integration is active. AI agent governance across the full stack is enabled by this partnership.
The four products that extend Cortex Sense across your stack
Permalink to “The four products that extend Cortex Sense across your stack”Enterprise Data Graph: 100+ connectors pull context from every data source (Salesforce, SAP, dbt, Looker, and 97 more) into a single living enterprise data graph with column-level lineage reverse-engineered from SQL. Cortex Sense reaches within the warehouse; the Enterprise Data Graph reaches across every system. See how it connects to Snowflake Cortex.
Context Agents: AI teammates that auto-generate descriptions, glossary terms, and ontology relationships from the Enterprise Data Graph. According to Atlan AI Labs, Context Agents have generated 690K+ descriptions, 87% rated on par or better than human writing across 50+ enterprise customers. These are the certified definitions Cortex Sense needs to exist in canonical form before it can enrich accurately.
Context Engineering Studio: Bootstraps, tests, and versions context as code. CI-integrated eval suites validate context before it reaches agents. The governed definitions CoWork and CoCo depend on are validated here before they reach runtime. One insurance customer compressed a one-year context build to one month using Context Engineering Studio. For teams asking how to build this foundation, context engineering is the discipline that makes AI accuracy repeatable at enterprise scale. The context engineering framework provides the structural approach teams need.
Context Lakehouse: Iceberg-native, API-accessible architecture. Open formats, graph and file architecture, vector-native AI search. The source that Atlan’s MCP server serves. Any agent (CoWork, CoCo, Cortex agents, or third-party agents) reads from the same governed source of truth.
The stack relationship: Challenge, Approach, Outcome
Permalink to “The stack relationship: Challenge, Approach, Outcome”Challenge: Cortex Sense performs best when the context it draws from is governed, consistent, and multi-system. Within Snowflake’s perimeter alone, signals are bounded by what Snowflake activity has generated.
Approach: Atlan layers on top of Snowflake (and Databricks, Purview, and 97 more connectors). It ingests Snowflake’s Horizon Context signals into the Enterprise Data Graph, enriches them with cross-system lineage and certified semantics, and serves them back to CoWork and CoCo via MCP server and A2A protocol. Teams looking to implement an enterprise context layer for AI can use this integration pattern as the starting point. The context layer for Snowflake guide covers the full native-plus-enterprise architecture in detail.
Outcome: Cortex Sense enriches agents with full enterprise context, not just Snowflake context. Atlan AI Labs measured a 5x accuracy improvement in agents grounded in Atlan’s context layer. As Baris Gultekin, VP of AI at Snowflake, said: “Context determines agent quality.” Atlan is where that context is built.
As analyst Mike Ni at Constellation Research (June 2026) noted, Snowflake’s Summit announcements center on “meaning, trust, permissions, context, and governance.” This is the exact foundation that semantic layer vs context layer thinking requires, and why the semantic layer alone is insufficient for enterprise AI.
Atlan works with CoWork and CoCo today. If your team is in the Cortex Sense private preview, talk to the Atlan team about connecting the Enterprise Data Graph as your cross-system context foundation.
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-BookReal stories from real customers: Context in production
Permalink to “Real stories from real customers: Context 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
Cortex Sense is the start of a larger context conversation
Permalink to “Cortex Sense is the start of a larger context conversation”Snowflake Cortex Sense is a genuine advance in enterprise AI. It automates the runtime context enrichment that used to require manual semantic model configuration, and the benchmark numbers (83% vs. 47%) are real and significant.
But the accuracy ceiling is set by the context Cortex Sense draws from. If that context is bounded to Snowflake’s perimeter, if definitions are inconsistent across systems, if lineage breaks at the transformation layer: Cortex Sense will enrich agents at scale with the wrong foundation. The agents will be confidently wrong.
The stack that closes this gap:
- Cortex Sense enriches Snowflake agents at runtime from Snowflake signals
- Atlan’s Enterprise Data Graph (100+ connectors) supplies the certified, cross-system foundation those signals need to be accurate
- Context Agents auto-generate the semantic layer CoWork and CoCo depend on (690K+ descriptions, 87% human quality)
- Context Engineering Studio validates that foundation before it reaches runtime
- Context Lakehouse delivers it to every agent via MCP, A2A, SQL, and open APIs
Cortex Sense + Atlan = agents with full enterprise context, not just Snowflake context. The stack is additive. The question for your team is not whether to use Cortex Sense. It is what foundation you are enriching it with.
FAQs about Snowflake Cortex Sense
Permalink to “FAQs about Snowflake Cortex Sense”-
Is Snowflake Cortex Sense available?
Cortex Sense is in private preview as of June 2, 2026. Snowflake has not specified a general availability date. Enterprises interested in early access should apply through Snowflake’s private preview program. (Source: Snowflake CoWork Press Release, June 2026) -
What is the difference between Cortex Sense and Horizon Context?
Horizon Context is the governance and metadata substrate: it manages semantic views, metadata connectors, and the business glossary. Cortex Sense is the runtime enrichment layer that activates Horizon Context signals for agents at query time. Horizon Context is the library; Cortex Sense is the librarian. (Source: Snowflake Horizon Context Blog, June 2026) -
What is the difference between Cortex Sense and Cortex Analyst?
Cortex Analyst is Snowflake’s natural language query interface for analysts: it translates questions into SQL against a semantic model. Cortex Sense is not a query interface. It is a runtime enrichment layer that enriches agents (CoWork and CoCo) with business context automatically, without requiring a manually configured semantic model per query. -
What signals does Cortex Sense use?
Cortex Sense builds context from four signal types: (1) query history (patterns from how analysts use data), (2) object metadata (table and column definitions), (3) BI dashboard definitions from Power BI and Tableau, and (4) Horizon Context semantic views (formal business definitions). (Source: Snowflake CoWork Blog, June 2026) -
Does Cortex Sense work with third-party tools like Tableau and Power BI?
Yes. Cortex Sense ingests metric and measure definitions from Power BI and Tableau dashboards as part of its signal collection. This makes existing BI definitions a source of enrichment context without requiring those definitions to be manually re-entered in Snowflake. (Source: Snowflake CoWork Blog, June 2026) -
What are Cortex Sense prebuilt plugins?
Cortex Sense ships with role-specific plugins for finance and sales. Each plugin bundles skills, business logic, and MCP connectors, enabling your team to deploy production-ready agents for those roles without rebuilding context from scratch. Additional role plugins are on the roadmap. (Source: Snowflake CoWork Press Release, June 2026) -
What is the difference between CoWork and Cortex Sense?
CoWork is the personal work agent for knowledge workers, the conversational interface. Cortex Sense is the context substrate that CoWork draws from at runtime. CoWork is the car; Cortex Sense is the navigation system that knows the roads, the speed limits, and the local rules of the road. -
How does Cortex Sense affect agent hallucinations?
By grounding agents in shared business definitions (revenue definitions, fiscal calendars, snapshot table logic), Cortex Sense reduces the conditions that produce AI agent hallucination. A hallucination often occurs when an agent lacks the business definition needed to answer accurately. Analyst David Linthicum (CIO.com, June 2026) noted Cortex Sense “could improve consistency, reduce the risk of hallucinations, and make AI outputs more operationally useful.”
Sources
Permalink to “Sources”- Snowflake CoWork Powers the Agentic Enterprise as the Personal Agent for Knowledge Workers, Snowflake
- Snowflake CoWork: The Personal Work Agent for Every Knowledge Worker, Snowflake Blog
- Snowflake Horizon Context: Governed Context for Enterprise AI, Snowflake Blog
- Snowflake moves up the AI stack, but the System of Intelligence is still being built, SiliconANGLE
- Snowflake recasts its AI strategy around action, not answers, with CoWork, CIO.com
- What Snowflake Summit 2026 signals about enterprise AI, InfoWorld
- Snowflake Summit 2026: Context, custom model training, Iceberg V3, Constellation Research
