What problem does the semantic layer solve?

The same question yields conflicting answers across different tables and metric definitions. For example, "active customer" might mean days_since_last_login 0 in another. The semantic layer curates one shared definition so all tools — dashboards, agents, and humans — query consistent data.

How much does it improve AI agent performance?

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. Agents also become more token-efficient when querying pre-aggregated data through semantic views, lowering inference cost.

Can semantic views be used outside Snowflake?

Yes. Snowflake made semantic views interoperable with Apache Ossie (incubating), the vendor-neutral open standard for semantic models, so users can read them in Snowflake with a single function call even if the models are stored elsewhere.

Large Language ModelsAI Business & IndustrySnowflake AI BlogAug 23, 2026Aug 23, 2026, 19:01 JST

Snowflake builds semantic layer to unify AI agents and analytics

Send on LINE

3 Key Points

What happened
Snowflake built an internal semantic layer — a standardized data interface sitting between raw tables and AI agents — to ensure consistent answers across its business. In July 2025 alone, more than 400 internal users ran over 5,400 queries through the product data science agent, and more than 5,600 employees across all teams 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 yields different answers depending on which table or metric definition is used. For example, "What is an active customer?" could mean days_since_last_login < 30, or credits_consumed > 0 with account_type != 'TRIAL', or a distinct count without a region identifier — creating conflicting answers and eroding trust in data. A semantic layer solves this by curating one definition for every interface, so both humans and AI agents query from the same authoritative source.
What to watch
Snowflake's semantic views increase text-to-SQL accuracy from 20% to more than 90% across 40 business questions in benchmark testing (per AtScale), and reduce agent token costs by enabling efficient use of pre-aggregated data. The company also made semantic views interoperable with Apache Ossie (incubating), the open standard for semantic models, allowing portability across platforms.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Get news like this every morning, free

Context & Analysis

Snowflake faced a familiar data governance challenge: petabytes of internal telemetry generated conflicting answers to the same business question. The root cause was not missing data but fragmented definitions — teams used different tables and metric calculations for concepts like "active customer," forcing analysts to play arbiter. As Snowflake scaled AI agents to query this data, the problem compounded: agents would report one number while legacy dashboards showed another, eroding confidence in both.

The company addressed this by building a semantic layer — a standardized logical abstraction sitting between raw tables and downstream consumers (dashboards, Cortex Agents, ad hoc queries). The semantic layer translates governed business language into physical schema, ensuring every interface sees the same meaning. In practice, this meant curating semantic views with consistent metric definitions, dimension descriptions, and synonyms that help AI agents understand what data means before they query it.

The results suggest the approach scales. In July 2025, Snowflake's product data science team fielded over 5,400 queries from more than 400 distinct users through a single semantic agent; company-wide, over 5,600 employees ran more than 320,000 queries through semantic layers spanning sales, HR, and support. Performance gains are material: benchmark testing showed text-to-SQL accuracy improved from 20% to more than 90% when semantic context was added, and agents became significantly more token-efficient by querying pre-aggregated materialized data. By versioning and testing semantic views using dbt integration and Snowflake's native continuous integration tools, Snowflake treated them as production code, enabling peer review and safe iteration.

FAQ

What problem does the semantic layer solve?: The same question yields conflicting answers across different tables and metric definitions. For example, "active customer" might mean days_since_last_login < 30 in one table or credits_consumed > 0 in another. The semantic layer curates one shared definition so all tools — dashboards, agents, and humans — query consistent data.
How much does it improve AI agent performance?: 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. Agents also become more token-efficient when querying pre-aggregated data through semantic views, lowering inference cost.
Can semantic views be used outside Snowflake?: Yes. Snowflake made semantic views interoperable with Apache Ossie (incubating), the vendor-neutral open standard for semantic models, so users can read them in Snowflake with a single function call even if the models are stored elsewhere.

Snowflake AI BlogRead Original Article

Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Get it free

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.