Enterprise AI agents have a new production failure mode, and it’s not a model. As enterprises move from a single-layer RAG to a hybrid search architecture, the same underlying data yields different answers depending on which agent, tool, or system is asking the question. Revenue means one thing in a business intelligence (BI) dashboard, something slightly different in an SQL table, and something else again in an agent manual. Building search infrastructure over the past two years has made vector search faster and cheaper. It did not create a shared definition of what the data meant.
At Snowflake Summit 26 in San Francisco, the data cloud vendor is addressing this challenge with announcements that include a Kafka-compatible managed streaming service called Data Stream, adaptive computing enhancements, expanded Apache Iceberg interoperability, and updates to its Cowork and CoCo agent and coding products. Underneath all of this is a context layer: Horizon Context and Cortex Sense, a two-layer system designed to provide agents with a shared business logic definition that is driven across search stacks. The context problem is why it matters: VentureBeat’s VB Pulse Q1 2026 data, Compiled from a survey of organizations with 100 or more employees, hybrid search intent rose from 10.3% in January to 33.3% in March, the fastest-growing strategic position in the database.
"There are many tools where you can ask a question, you get a very confident answer, but whether it is true or not is different," said Christian Kleinerman, EVP of product at Snowflake.
From fragmented business logic to a managed context layer
Horizon Context targets are problem specific. Today, business logic is distributed across SQL, BI dashboards, and agent instructions, and no single system has this definition. When multiple agents or tools query the same underlying information, they think of different schemas and return different responses. Horizon Context is Snowflake’s attempt to fix this at the directory layer, not the agent layer.
Horizon Context. A customer-managed layer built upon Snowflake’s acquisition of Select Star. It pulls metadata from Postgres, SQL Server, Tableau, and Power BI into Horizon Catalog, so every agent, BI tool, and external system uses the same managed definition instead of independently thinking about the raw physical schema. Semantic View Autopilot automatically creates and refines semantic views over time, extending selected business logic without requiring continuous manual effort.
Sense of the cortex. A layer derived from the platform. It automatically builds and enriches context from customer data and usage patterns on an ongoing basis without requiring a manual semantic view author. Kleinerman described this as improving the default experience before any overt curation takes place.
The difference between the two layers is architecture, and Kleinerman was clear about that. "Think of Horizon Context as everything exposed and declared by clients, and Cortex Sense as everything hidden and accessed by us." Kleinerman said.
The two layers connect to Snowflake’s existing search infrastructure. Cortex Search, the company’s RAG application, connects to both CoCo and Cowork as a tool, so contextual search enriched by both layers flows into workflows.
Although Horizon Context is a Snowflake technology, the goal is to make it interoperable and open. Snowflake links the technology to the Open Semantic Exchange, making client-declared definitions portable between third-party catalogs and tools.
"Horizon Context, we are 100% committed and strive to make sure it is not locked," Kleinerman said.
Context layers are everywhere. The question is which ones actually work.
Snowflake joins an increasingly crowded field of vendors targeting the same problem. Microsoft opened Fabric IQ business ontology via MCP thus, any vendor’s agent can draw from the shared semantic layer. Redis launched Irisa context and storage platform that sits between agents and their data, built on a redesigned storage engine for agent-scale search volumes. Pinecone embeds vectors from a database into a knowledge engine With Nexus, which aggregates enterprise data into task-specific artifacts before agents query them.
Devin Pratt, director of research at IDC, told VentureBeat that he thinks Snowflake is going in the right direction and going where the rest of the market is going.
"Agents are only as good as the data and semantics behind them, so the context layer, not the model, is the thing to watch right now," Pratt said.
According to Pratt, what works in Snowflake’s version is fragmentation. Horizon Context covers what teams declare and curate themselves, while Cortex Sense covers what the platform automatically picks up. Equally important, they put Horizon Context inside the directory and management layer instead of locking it in after the fact.
"The context layer is the real battleground for agent AI. An agent is only as reliable as the data and semantics behind it" Pratt said.
Mike Leone, vice president and principal analyst at Moor Insights and Strategy, agreed that treating the two layers differently is the right architectural call.
"I like where Snowflake is going. They divide their context into two buckets, with Horizon Context, which covers what customers explicitly define, and Cortex Sense, which covers what the platform itself understands." Leone told VentureBeat. "You can’t trust these two things the same way, so treating them differently is the right call. If Snowflake can show that these two layers align neatly, and you can see where each answer comes from, then they have something real."
What this means for businesses
For enterprises that value context layers, the architectural direction is clear. Not an implementation gap.
Agents are raising the bar on an age-old problem. The idea of a semantic layer has been around for years, but agents change the cost of failure—when an agent makes a wrong response at scale, the damage is immediate. Leone is straightforward about what this means for most sellers on the market right now.
"Most sellers who sell a pop-up fix are over-promising," Leon said. "Drop one into a real enterprise and it basically reveals how messed up your data and definitions are, and many companies are about to find that out the hard way."
The rating bar is specific. Pratt identified what separates working context layers from persistent ones: governance and generation built so that teams can verify why an agent is responding, portability, context and policy independent of a single vendor, and accuracy that can be measured and reused across agents and tools.
"Enterprises don’t need another silo of semantics," Pratt said. "They need a context layer that is manageable, portable, and reliable enough to audit."
