The data definition gap: Why your AI agent sounds confident but gets it wrong

Snowflake has introduced a two-layer context system designed to stop AI agents from returning contradictory answers when querying the same business data. Sean Michael Kerner reports for VentureBeat that the announcements came at Snowflake Summit 26 in San Francisco and center on two new products: Horizon Context and Cortex Sense.

The core problem is that business logic is scattered across SQL tables, BI dashboards and agent instructions. The word “revenue,” for example, can mean different things depending on which system interprets it. When multiple agents query the same underlying data, they reason over different definitions and return different results. Christian Kleinerman, EVP of Product at Snowflake, put it plainly: “There are a lot of tools out there that you can ask questions, you get a very confident answer, but whether it’s correct or not is different.”

Two layers, two jobs

• Horizon Context covers what customers explicitly define. Built on Snowflake’s acquisition of Select Star, it pulls metadata from sources including Postgres, Tableau and Power BI into a shared catalog. A feature called Semantic View Autopilot refines these definitions over time without manual effort.
• Cortex Sense covers what the platform derives automatically from usage patterns, without requiring any manual configuration.

Kleinerman described the split clearly: “Think of Horizon Context as everything that is explicit and declared by customers, and Cortex Sense is anything that is implicit and derived by us.”

Snowflake is not alone in targeting this problem. Microsoft, Redis and Pinecone are all developing similar context layers for enterprise AI. Devin Pratt, research director at IDC, told VentureBeat: “The context layer is the real battleground for agentic AI. An agent is only as trustworthy as the data and semantics behind it.”

Analysts broadly support the architectural direction but caution that execution is harder than it looks. Mike Leone of Moor Insights and Strategy warned that most vendors are overpromising: “Drop one into a real enterprise and it mostly exposes how messy your data and definitions already are.” Snowflake says Horizon Context is built on an open standard called Open Semantic Interchange, making definitions portable across third-party tools.