How many internal queries went through Snowflake's semantic layer in July 2025?

In July 2025 alone, more than 400 distinct internal users ran over 5,400 queries through the product data science agent and semantic layer. Across all Snowflake teams and agents, more than 5,600 internal employees ran over 320,000 queries using the broad semantic layer in sales, HR, support and other areas.

What improvement did semantic context bring to text-to-SQL accuracy?

In AtScale's benchmark testing, adding semantic context increased text-to-SQL accuracy from 20% to more than 90% across 40 business questions on TPC-DS.

Large Language ModelsAI Business & IndustrySnowflake AI Blog3d agoAug 18, 2026, 06:00 JST3 min read

Snowflake builds internal semantic layer to unify AI agent queries

Send on LINE

Key takeaway

Snowflake has built an internal semantic layer—a standardized business logic translation system—that sits between raw data and AI agents, allowing consistent query results across the entire company.
In July 2025, over 5,600 internal employees ran more than 320,000 queries through this semantic layer across sales, HR, support and other teams, with one benchmark showing semantic context increased text-to-SQL accuracy from 20% to more than 90%.
The layer reduces token costs for AI agents and eliminates the need for manual query reconciliation.

3 Key Points

What happened
Snowflake created an internal semantic layer—a standardized data translation system sitting between raw database tables and downstream tools—to give AI agents and business users consistent, governed access to corporate data. In July 2025 alone, more than 400 internal users ran over 5,400 queries through the product data science agent, and across all Snowflake teams, more than 5,600 employees ran over 320,000 queries using the broader semantic layer in sales, HR, support and other areas.
Why it matters
Without a shared semantic layer, the same question (such as "What is an active customer?") yields conflicting answers across different data tables and metric definitions. By curating one consistent meaning for data through semantic views, Snowflake reduced the burden on data science teams to manually reconcile numbers and gave AI agents two immediate wins: faster query execution (agents no longer waste time sampling multiple raw sources) and lower token costs (pre-aggregated data reduces repeated computation). In one benchmark, adding semantic context increased text-to-SQL accuracy from 20% to more than 90% across 40 business questions.
What to watch
Snowflake's semantic views are interoperable with Apache Ossie™ (incubating), a vendor-neutral open standard, allowing the models to work across platforms. The company also uses Snowflake CoCo (an AI assistant) to draft and refine semantic processes, and Cortex Agents automatically scan across semantic views using custom routing instructions to find the right data context for a given query.

Ask the AI about this article →

Summaries like this, in your inbox every morning.

Get news like this every morning — free

Context & Analysis

Snowflake's internal data challenge mirrors a widespread enterprise problem: massive volumes of telemetry and business data do not automatically produce reliable answers. The article illustrates this by showing how a single question—"What is an active customer?"—yields three different answers depending on which table and metric definition is consulted. Without semantic alignment, the burden falls on data science teams to manually verify queries and reconcile conflicting numbers, a process that becomes untenable when new AI agents enter the picture.

The semantic layer solves this by introducing a "golden layer" that translates raw physical tables into consistent business concepts (dimensions, metrics, and facts) before exposing them to downstream consumers—dashboards, Cortex Agents, and ad hoc users. This approach yields compounding returns: AI agents run faster because they query pre-structured, pre-aggregated data rather than raw tables; token efficiency improves because agents no longer repeat computation; and accuracy jumps dramatically (20% to over 90% in one benchmark) because the model receives precise business context instead of raw column names. Snowflake's own July 2025 usage numbers—over 320,000 queries across 5,600+ employees—validate that the layer is functional and trusted at scale.

FAQ

How many internal queries went through Snowflake's semantic layer in July 2025?: In July 2025 alone, more than 400 distinct internal users ran over 5,400 queries through the product data science agent and semantic layer. Across all Snowflake teams and agents, more than 5,600 internal employees ran over 320,000 queries using the broad semantic layer in sales, HR, support and other areas.
What improvement did semantic context bring to text-to-SQL accuracy?: In AtScale's benchmark testing, adding semantic context increased text-to-SQL accuracy from 20% to more than 90% across 40 business questions on TPC-DS.

Snowflake AI BlogRead Original Article