The AI agent will continue to get the wrong answer with confidence. The context layer is the next production problem for enterprise AI.

Enterprise AI agents have a new production failure mode, and this is not the model. As organizations move from single-layer RAG architectures to hybrid recovery architectures, the same underlying data produces 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 a SQL table, and yet another thing in an agent statement. The establishment of retrieval infrastructure over the past two years has enabled faster and less expensive vector search. It did not result in a common definition of the meaning of the data.

At Snowflake Summit 26 in San Francisco, the data cloud provider is tackling this issue in a big way, with announcements covering a Kafka-enabled managed streaming service called Data Stream, improvements to adaptive computing, expanded Apache Iceberg interoperability, and updates to its Cowork and CoCo agents and coding products. Underneath it all is a layer of context: Horizon Context and Cortex Sense, a two-layer system designed to give agents a governed, shared definition of business logic across recovery stacks. The context issue is why this matters: VentureBeat's VB Pulse Q1 2026 data, drawn from a survey of organizations with 100 or more employees, shows that hybrid recovery intent tripled from 10.3% in January to 33.3% in March, the fastest-growing strategic position in the data set.

“There are a lot of tools for asking questions and you get a very confident answer, but whether it's correct or not is different,” said Christian Kleinerman, executive vice president of product at Snowflake.

From fragmented business logic to a governed context layer

The problem Horizon Context targets is specific. Today, business logic is split between SQL, BI dashboards, and agent statements, and no single system has the definition. When multiple agents or tools query the same underlying data, they reason about different patterns and return different answers. Horizon Context is Snowflake's attempt to solve this problem at the catalog layer rather than the agent layer.

Skyline background. The customer-managed layer, based on Snowflake's acquisition of Select Star. It pulls metadata from Postgres, SQL Server, Tableau, and Power BI into the Horizon catalog, so that each agent, BI tool, and external system relies on the same governed definition rather than reasoning independently about a raw physical schema. Semantic View Autopilot automatically creates and refines semantic views over time, extending organized business logic without requiring ongoing manual effort.

Sense of cortex. The platform derived layer. It automatically creates and enriches context from customer data and usage patterns on an ongoing basis, without requiring manual creation of semantic views. Kleinerman described it as improving the default experience before explicit curation takes place.

The distinction between the two layers is architectural and Kleinerman was specific about it. “Think of Horizon Context as anything that is explicit and stated by customers, and Cortex Sense is anything that is implicit and derived by us,” Kleinerman said.

Both layers connect to Snowflake's existing recovery infrastructure. Cortex Search, the company's RAG implementation, connects to both CoCo and Cowork as a tool, so context enriched by either layer flows into retrieval workflows.

Although Horizon Context is a Snowflake technology, the goal is for it to be interoperable and open. Snowflake combines the technology with open semantic exchange, making customer-declared definitions portable between catalogs and third-party tools.

“Context Horizon, we are 100% committed and leading the effort to make sure this is not locked in,” Kleinerman said.

Layers of context are everywhere. The question is which ones actually work.

Snowflake joins a growing group of vendors targeting the same problem. Microsoft has opened its Fabric IQ business ontology via MCP so that any vendor's agent can rely on a shared semantic layer. Redis launched Iris, a context and memory platform between agents and their data, built on a redesigned storage engine for agent-scale retrieval volumes. Pinecone is moving from a vector database to a knowledge engine with Nexus, which compiles business data into task-specific artifacts before agents query them.

Devin Pratt, research director at IDC, told VentureBeat that in his view, Snowflake is moving in the right direction and going where the overall market is heading.

“Agents are only as good as the data and semantics behind them, so it’s the context layer, not the model, that is the thing to watch right now,” Pratt said.

According to Pratt, what works in Snowflake's version is the split. Horizon Context covers what teams report and organize themselves, and Cortex Sense covers what the platform automatically retrieves. Just as importantly, they anchored Horizon Context into the catalog and governance layer rather than adding it as an afterthought.

"The contextual layer is the real battlefield of agentic 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 decision.

"I like the direction Snowflake is taking. They divide context into two categories, Horizon Context covering what customers explicitly define and Cortex Sense covering what the platform discovers on its own," Leone told VentureBeat. "You can't trust these two things the same way, so treating them differently is the right decision. If Snowflake can show that these two layers reconcile cleanly and you can see where each response is coming from, there's something real about them."

What this means for businesses

For companies evaluating context layers, the architectural direction is clear. The execution gap is not.

Agents are raising the bar on an old problem. The idea of ​​the semantic layer has been around for years, but agents change the cost of failure: when an agent gives a wrong answer on a large scale, the damage is immediate. Leone clearly explains what this means for most providers on the market today. “Most sellers offering an instant fix overpromise,” Leone said. “Put one in a real company and it mostly reveals how complicated your data and definitions already are, and many companies are about to find out the hard way. »

The rating bar is specific. Pratt identified what differentiates contextual layers that work from those that stagnate: built-in governance and lineage so teams can verify why an agent gave the response, portability so context and policy aren't limited to a single provider, and accuracy that can be measured and reused across agents and tools.

“Businesses don’t need another semantic silo,” Pratt said. “They need a layer of context that is sufficiently governed, portable, and reliable to perform an audit on.”