Sridhar said the model is only as good as the context.
Atlan is how Snowflake on Snowflake gets there.
Cortex Agents need semantic models, lineage, and human-curated meaning to act. Atlan is the open context layer that sits on top of Cortex, Polaris, and Horizon — and feeds Snowflake's own AI Data Cloud first.
Enterprise AI fails not because of the model, but because of missing context
The wall isn't the models. It's that no agent can reason effectively about a business it doesn't understand — what your data means, how your teams work, how your company defines "revenue" compared to the rest of the world.
Key Insight
When every organization has access to the same intelligence, context becomes the differentiator. The enterprise that best articulates its own knowledge — its data, its processes, its meaning — will build AI that's most useful to its people.
"We built a revenue analysis agent and it couldn't answer one question. We started to realize we were missing this translation layer. We had no way to interpret human language against the structure of the data."
Joe DosSantos
VP, Enterprise Data & Analytics
Cortex has the runtime.
The shared meaning isn't in it yet.
Three things are true at Snowflake right now. Cortex Agents went GA in June 2025. Snowflake Intelligence is live in preview. Sunny Bedi's team runs Snowflake on Snowflake — dozens of internal data domains, hundreds of agents in development, thousands of semantic definitions today scattered across YAML files in source control, Confluence pages, and engineering heads.
The semantic model spec inside Cortex Analyst is the right primitive. It defines tables, dimensions, measures, synonyms — the meaning an LLM needs before it generates SQL. But the spec is a file. It doesn't carry lineage upstream to the warehouse columns. It doesn't roll forward when a dbt model changes. It doesn't expose which version is canonical when three teams ship their own.
Snowflake has set the bar for the rest of the industry: agents in production, on warehouse data, with governance. The bar is now what Snowflake's own data org has to clear. Without a context layer between Polaris, Horizon, and Cortex, every Snowflake team rebuilds the same understanding from scratch, every agent answers from a partial slice, and the agent runtime becomes only as good as the YAML someone remembered to update.
Question It Raises
Context Layer
Answer It Needs
Who's asking — and what decision?
CS team or Sales team?
CS team optimizes for renewal risk
What does "customer" mean here?
Account or individual?
Parent account, not individual location
How do you define "top"?
Revenue, orders, or margin?
Top = highest net ACV, not order count
Which tables hold net ACV?
CRM vs. billing?
Use billing.subscriptions joined with crm.accounts
How do you calculate revenue?
Gross or net of discounts?
Revenue net of discounts and refunds
No new warehouse. No fork of Cortex. The context layer ships in weeks.
Four moves between Snowflake's current stack and an agent that knows the AI Data Cloud business. Cortex, Polaris, Horizon, Snowpark, Snowflake Notebooks, and the dbt models on top all stay where they are. Atlan binds the meaning and pushes it back into Cortex through the Snowflake MCP server.
MAP
Map every Snowflake account, schema, and Iceberg table
Atlan's Snowflake connector reads from Polaris Catalog, Horizon, Snowpipe Streaming, dbt, and the Cortex Analyst semantic models. Lineage is drawn column by column across every Snowflake account inside Snowflake — finance, GTM, product analytics, AI Lab. The map shows up in days, not quarters.
"Within the first year we cataloged over 18 million assets, defined more than 1300 glossary terms. Atlan had lineage across our on-prem Oracle databases, BigQuery, and Looker."
Kiran Panja
Managing Director, Cloud & Data Engineering
BIND
Bind business meaning to the Cortex semantic model
Definitions for ARR, consumption credits, customer, product edition, and the fifty other terms that appear in every Snowflake board deck land in one place — owned by the people who set them. The Snowflake-specific meaning that already lives in Sunny Bedi's data org becomes the meaning Cortex Analyst reads before it generates SQL. The Confluence page stops being the source of truth. Atlan is.
"We're focused on how we can scale context development as much as possible, and where can we leverage Atlan AI to build the most robust definitions across our data estate."
Takashi Ueki
Head of Enterprise Data & Analytics
GOVERN
Govern who and what can read the context
Horizon's masking, classification, and access history connect to Atlan's approval flows and lineage proofs. Every Cortex Agent read carries the human source of the definition it acted on, the team that certified it, and the policy that allowed it. Auditable on every call, not on a quarterly review.
"Atlan gives us a UI that our community can use to edit, update and manage classifications as well as other metadata enrichments into a verified state."
Sherri Adame
Enterprise Data Governance Leader
ACTIVATE
Hand the context to every agent in the Cortex runtime
Cortex Analyst, Cortex Search, Cortex Agents, and Snowflake Intelligence all read the same context through Atlan's MCP server. Trace which definition any answer came from. Promote what works. Retire what doesn't. The same loop your customers want — you run it on yourselves first.
"All of the work that we did to get to a shared language amongst people at Workday can be leveraged by AI via Atlan's MCP server."
Joe DosSantos
VP, Enterprise Data & Analytics
The architecture choice that
outlives the model underneath it
Snowflake is multi-cloud, multi-model, and pre-A2A. The context layer can't be a bolt-on to one runtime. These are the three commitments Atlan makes to Christian's product org.
Open and portable
The catalog choice can't lock you into one runtime. Atlan is open by design — Apache Iceberg-native, MCP-first, runs against Polaris, Horizon, Snowpark, Streamlit, and whatever Cortex ships next. Polaris is open. The context that points at Polaris should be too.
AI-Native
Catalogs were built for humans browsing schemas. Atlan was built for agents reading data. Lineage, definitions, and governance render as machine-readable context that Cortex Analyst, Cortex Agents, and Snowflake Intelligence consume directly through MCP — with traceability back to the human source.
Humans of data run the system
Sunny Bedi's data org owns the meaning. Baris's AI org owns the runtime. Jeff Hollan's team owns the agent experience. Atlan is the workflow they govern in — not a separate tool, not a separate team. The context layer lives wherever the business meaning is already being decided.
A leader across every context category
"The Metadata Lakehouse forms the core foundation, built on an open and highly performant architecture. It is designed to be Iceberg-native and includes a knowledge graph for business domains, vector storage, and analytics, which is purpose-built for AI."
Leader in the 2025 Gartner® Magic Quadrant™ for Metadata Management Solutions
Read the Gartner MQ report
"Atlan stands out in AI-native governance through context-based partnerships, agentic stewardship and orchestration of enterprise agentic systems. They take a partnership and co-innovation based approach, which is reflected in their App Framework as a marketplace for context."
Leader in the 2026 Gartner® Magic Quadrant™ for Data & Analytics Governance
Read the Gartner D&A report
A Leader and a Customer Favourite in the Forrester Wave™
Data & Analytics Governance Solutions and Enterprise Data Catalogs
Context is IP.
Keep yours.
