The enterprise context stack has two layers. The data plane holds your data, runs your queries, and now, with Horizon Context, enriches that data with Snowflake-native semantic definitions — what Snowflake calls its context layer for Snowflake Cortex. The context plane holds the cross-system intelligence that AI agents need to act on data across every tool, warehouse, and business domain — far beyond what a semantic layer alone can provide. Snowflake builds the first. Atlan builds the second. In production, they work together.
What Snowflake Horizon Context does
Permalink to “What Snowflake Horizon Context does”Snowflake Horizon Context is a component of Snowflake Horizon Catalog, announced June 2, 2026 at Snowflake Summit. It follows a collect / enrich / activate framework: pull signals from external data sources, enrich them with business definitions, and deliver the result to Snowflake’s AI agents.
As Christian Kleinerman, EVP of Product at Snowflake, described it at the Summit keynote: Horizon Context “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.”
What it ships as of June 2026:
- Metadata Connectors (Wave 1, private preview): 5 external connectors for PostgreSQL, Microsoft SQL Server, Tableau, Power BI, and dbt. These pull schemas, query logs, and dashboard definitions into Horizon’s semantic layer. Snowflake’s architecture roadmap targets broader coverage; Wave 1 is the committed launch set.
- Semantic Studio (private preview): AI-assisted IDE inside Snowflake Workspaces for defining business logic without writing SQL. Git-based versioning, CoCo integration, collaborative metric authoring.
- Semantic View Autopilot (available): Auto-generates semantic views from existing SQL, Tableau workbooks, or Power BI reports. A genuine time-saver for Snowflake-native teams.
- AI-Generated Documentation (available): Auto-generates table and column descriptions from metadata and optional sample data. Static output, Snowflake-perimeter only.
- Business Glossary: H2 2026 roadmap. Not yet shipped.
Where Horizon Context delivers real value: for organizations running primarily Snowflake-native AI workloads, it creates a consistent semantic foundation within the Snowflake perimeter. Cortex Sense operating on Horizon Context reaches 86% accuracy on structured questions, compared to 24% with frontier models alone. That accuracy gap, closed by semantic enrichment, is the clearest argument for what context does for AI.
What it does not yet cover: cross-system semantics beyond the 5 Wave 1 connectors, a shipped Business Glossary, and context enrichment that compounds over time rather than generating static output once.
See Context Studio in Action
Context Engineering Studio is Atlan's GA environment for bootstrapping, testing, and versioning context as code. Watch the live demo to see how it works alongside Snowflake's semantic layer.
Watch the DemoWhat Atlan’s context layer adds on top
Permalink to “What Atlan’s context layer adds on top”Atlan’s enterprise context layer does not replicate what Horizon Context does. It extends it, functioning as the agent context layer that delivers cross-system knowledge. Atlan ingests Horizon Context metadata as one data source among many, adds the cross-system coverage that Horizon cannot yet reach, and serves the enriched result back to the same Snowflake agents Horizon feeds.
Four products do this work:
Enterprise Data Graph (100+ connectors, GA). Ingests Horizon Context as one source among 100+, adding BigQuery, Redshift, Salesforce, Looker, MongoDB, Databricks, legacy warehouses, and more. Column-level lineage across 80+ systems gives AI agents the provenance they need to trust answers from complex multi-hop pipelines. Every connector is generally available today.
Context Engineering Studio (GA). Atlan’s context engineering framework environment for bootstrapping, testing, and versioning context as code. CI-integrated eval suites validate context before it reaches any agent using context engineering for AI agents principles — the same way code is tested before deployment. One insurance enterprise compressed a one-year context build into one month using it.
Context Agents (GA). Autonomous agents that generate and enrich metadata descriptions across all connected systems. 690,000+ descriptions generated across 50+ enterprises by April 2026, with 87% rated on par with or better than human-written content. Unlike Horizon’s static AI-generated documentation, Context Agents compound: each enrichment pass builds on the last, growing the enterprise knowledge base over time.
Context Lakehouse. Stores enriched context in Iceberg-native open formats with vector-native AI search. Queryable by any agent via MCP, A2A protocol, SQL, or APIs — making it a true agent context layer. Not bounded by the Snowflake warehouse architecture, which means any agent, from any tool, can retrieve the same unified context.
The combined result: agents grounded in Atlan’s full context graph-powered context layer deliver a 5x accuracy improvement compared to agents operating on raw metadata alone (Atlan AI Labs). That stat applies across the full enterprise stack, not just within Snowflake.
The capability comparison
Permalink to “The capability comparison”The table below maps each key dimension of Horizon Context against the Atlan context layer. Both systems are actively evolving; the status reflects June 2026.
| Capability | Snowflake Horizon Context | Atlan context layer | Notes |
|---|---|---|---|
| Metadata connectors | 5 (Wave 1, private preview) | 100+ (GA) | Horizon architecture diagram claims more; Wave 1 is the committed set |
| Business Glossary | H2 2026 roadmap, not shipped | Shipped | Closes agents’ definitional ambiguity today, not later |
| Context agents / auto-documentation | AI-generated docs (static, Snowflake-only) | Context Agents: 690K+ descriptions, 87% quality, 50+ enterprises, cross-system | Atlan’s agents compound; Horizon’s docs are one-time enrichment |
| Semantic authoring tool | Semantic Studio (private preview) | Context Engineering Studio (GA) | CES adds CI eval suites, versioning, and context-as-code pipeline |
| Cross-system coverage | Snowflake perimeter + 5 Wave 1 connectors | 100+ systems: BigQuery, Redshift, Salesforce, Looker, MongoDB, legacy | Horizon cannot reach Salesforce, legacy systems, or most cloud warehouses today |
| Lineage | Tabular-level external lineage (Snowflake-native) | Column-level lineage across 80+ systems | Column-level lineage required for agent trust in complex pipelines |
| Context delivery to agents | Cortex, CoWork, CoCo via Horizon Catalog | Any agent via MCP, A2A, SQL, APIs, including Cortex, CoWork, CoCo | Atlan is Snowflake’s 2025 Data Governance Partner of the Year |
| Agent accuracy improvement | Cortex Sense: 86% vs 24% baseline (Snowflake-only queries) | 5x accuracy improvement across full enterprise stack (Atlan AI Labs) | Snowflake stat applies within Snowflake; Atlan stat applies across all systems |
| Context store architecture | Semantic views within Snowflake warehouse | Context Lakehouse: Iceberg-native, open formats, vector-native AI search | Atlan’s context store is designed for cross-system AI retrieval, not warehouse-bounded |
The business glossary gap is the most operationally significant row in this table. Without consistent terminology, AI agent hallucination risk increases. Without consistent terminology, AI agent hallucination risk increases. AI agents need consistent definitions to give consistent answers. Without a Business Glossary, the same metric queried by a finance team and a product team can return different answers. Horizon Context’s Business Glossary arrives H2 2026. Atlan’s is shipped today.
The cross-system coverage gap matters most for enterprises that do not run on Snowflake alone. If your AI agents need to reason across Salesforce CRM data, Looker metrics, and Snowflake warehouse data in a single query, five Wave 1 connectors are not enough. That is the production reality for most enterprise data teams. An enterprise data graph connecting all those systems is the architectural answer. An enterprise data graph connecting all those systems is the architectural answer.
Download the CIO Guide to Context Graphs
The four-layer context graph architecture for enterprise AI, covering the metadata foundation, semantic enrichment, context delivery, and agent orchestration layers. Includes implementation steps for Snowflake-primary data estates.
Download the GuideReal stories from real customers: Context layer in production
Permalink to “Real stories from real customers: Context layer 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 and 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 and Analytics Officer, DigiKey
How they work together in practice
Permalink to “How they work together in practice”The integration architecture has five steps.
Challenge: An enterprise runs AI agents that query across Snowflake, Salesforce, and Looker. Horizon Context enriches the Snowflake data with semantic definitions. But the Salesforce CRM data and the Looker dashboard metrics have no semantic enrichment, no business glossary alignment, and no lineage connecting them back to the Snowflake tables the agents query. The agents return inconsistent answers.
Approach:
- Horizon Context enriches Snowflake metadata with business definitions via its collect / enrich / activate pipeline
- Atlan ingests the Horizon Context enriched metadata via MCP as one input to the Enterprise Data Graph
- Atlan adds Salesforce, Looker, and all other connected systems to the same graph, with column-level lineage connecting them
- Context Agents autonomously generate and refine business descriptions across all connected systems, compounding over each enrichment cycle
- Atlan’s MCP server delivers the unified enterprise context back to Snowflake’s Cortex agents, CoWork, and CoCo agents
Outcome: Agents that previously had Snowflake-perimeter context now have enterprise context. The Salesforce opportunity stage, the Looker revenue metric, and the Snowflake orders table share a common semantic foundation. Consistent answers across all queried systems, with AI agent governance over what context each agent can access. The context engineering methodology that powers this is documented in Atlan’s field guide, and the context layer implementation guide walks through the step-by-step architecture with Snowflake as the data plane.
Why Snowflake and Atlan are stronger together
Permalink to “Why Snowflake and Atlan are stronger together”Horizon Context is a genuine step toward the enterprise context layer. Its semantic enrichment within the Snowflake perimeter is real and useful. Cortex Sense’s 86% accuracy improvement, earned with Horizon Context enrichment, is the clearest evidence of what context does for AI. The governance enforcement at the meaning level, not just the table level, is something few catalog products have historically delivered.
The production deployments that reach scale share a common characteristic: context that spans the full enterprise stack. That means a Business Glossary with stewardship workflows, cross-system lineage at the column level, context agents that compound knowledge over time, and a context store that serves any agent regardless of where the data lives. Horizon Context builds toward this. Atlan delivers it today, and ingests from Horizon Context to extend Snowflake’s own enrichment enterprise-wide.
At Workday, Nasdaq, Mastercard, General Motors, and 50+ other enterprises, this combination is how AI deployments reach production rather than stalling in pilot. Snowflake builds and enriches the data plane. Atlan builds and maintains the context plane.
FAQs about Snowflake Horizon Context and Atlan
Permalink to “FAQs about Snowflake Horizon Context and Atlan”1. How do Snowflake Horizon Context and Atlan work together?
Snowflake Horizon Context enriches Snowflake metadata with semantic definitions via its collect / enrich / activate pipeline. Atlan ingests that enriched metadata via the context layer for Snowflake integration as one input to the Enterprise Data Graph, adds cross-system coverage from 100+ connected systems, and serves the unified enterprise context back to Snowflake’s Cortex agents, CoWork, and CoCo via Atlan’s MCP server. The architectural relationship: Snowflake is the data plane, Atlan is the context plane.
2. What does Atlan add that Snowflake Horizon Context does not have?
As of June 2026: 100+ GA connectors versus Horizon’s 5 in private preview; a shipped Business Glossary versus Horizon’s H2 2026 roadmap; Context Agents with 690K+ autonomous descriptions across 50+ enterprises versus static AI-generated docs; column-level lineage across 80+ systems versus tabular-level Snowflake-native lineage; and a Context Lakehouse in Iceberg-native open formats queryable by any agent via MCP.
3. Does Atlan replace Snowflake Horizon Context?
No. Atlan extends it. Atlan treats Horizon Context as a data source, ingesting Snowflake’s enriched metadata into the Enterprise Data Graph and adding cross-system coverage. The relationship is additive: Horizon Context enriches the Snowflake perimeter; Atlan extends that enrichment enterprise-wide.
4. Can Atlan read from Snowflake Horizon Context?
Yes. Atlan ingests Horizon Context metadata via the MCP standard that Horizon Catalog exposes for external agents. This makes Snowflake’s enriched metadata available to Atlan’s Enterprise Data Graph, where it joins cross-system lineage and business definitions from 100+ other connected systems.
5. What is the biggest capability gap between Horizon Context and an enterprise context layer?
Two gaps are most operationally significant: Business Glossary (ships H2 2026 in Horizon; already shipped in Atlan) and cross-system connector coverage (5 in Horizon Wave 1; 100+ GA in Atlan). Business Glossary determines whether AI agents give consistent answers across business units. Connector coverage determines whether agents can reason across all your systems. Both gaps are addressed by adding Atlan’s context layer on top of Snowflake Horizon Context.
6. Is the “Snowflake = data plane, Atlan = context plane” framing official?
It reflects Atlan’s published architectural position and is consistent with Snowflake’s own categorization of Atlan as its 2025 Data Governance Partner of the Year. The complementary architecture is running in production at Workday, Nasdaq, Mastercard, General Motors, and 50+ other enterprises. For the full architectural framing, see the context layer for Snowflake native + enterprise guide, with the context engineering framework as the methodological companion.
Sources
Permalink to “Sources”- Snowflake Horizon Context, Official Product Page, Snowflake
- Snowflake Horizon Context: The Governed Context Layer for AI, BI and Apps, Snowflake Blog
- Snowflake Advances Trusted AI with Snowflake Horizon Catalog, Snowflake Press Release
- Snowflake moves up the AI stack, but the System of Intelligence is still being built, SiliconANGLE
- What Is an Enterprise Context Layer? A Field Guide for AI Teams, Atlan
- Context Layer for Snowflake: Native + Enterprise Guide 2026, Atlan
- Context Agents, Atlan
- What is Snowflake Horizon Context?, Atlan
- Context Engineering: The Framework That Makes AI Agents Production-Ready, Atlan
- How to Implement an Enterprise Context Layer for AI, Atlan
