Snowflake has built an internal semantic layer that acts as a single authoritative source of business meaning, allowing AI agents and analytics tools to query consistent data without manual reconciliation.
In July 2025, the semantic layer handled over 320,000 queries from more than 5,600 internal employees.
By standardizing data context this way, Snowflake increased text-to-SQL accuracy from 20% to over 90% in benchmark testing, and agents achieve faster execution and lower token costs by avoiding repeated joins and raw table sampling.
What happened
Snowflake created an internal semantic layer—a standardized data interface that sits between raw tables and downstream consumers (dashboards, AI agents, business intelligence tools)—to provide consistent definitions and context. In July 2025 alone, over 400 internal users ran more than 5,400 queries through the product data science agent using this semantic layer; across all teams, more than 5,600 employees ran over 320,000 queries via the broader semantic layer in sales, HR, support and other areas.
Why it matters
Without a shared semantic layer, the same business question (e.g., "What is an active customer?") yields conflicting answers from different data tables and metric definitions. By curating one authoritative semantic view, Snowflake ensures that both humans and AI agents query the same trusted data, eliminating the need for data scientists to manually reconcile discrepancies. In AtScale's benchmark testing, 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 materialized (pre-aggregated) automatically to reduce query latency and token usage for AI agents. The semantic views are version-controlled and tested like production software, and are interoperable with Apache Ossie (incubating), a vendor-neutral open standard—meaning businesses can use their semantic models across platforms and tools.
Ask the AI about this article →
Snowflake's challenge was not scarcity of data—the company collects petabytes of internal telemetry across every query, warehouse, and user click—but clarity. When millions of data sources exist without shared definitions, the raw volume becomes a liability: different teams, dashboards, and tools derive conflicting answers from the same underlying facts. The company recognized that this alignment problem compounds when AI agents enter the picture. Because new metrics and data sources are built rapidly, information gaps emerge even for skilled analysts; if an AI agent reports one number while a legacy dashboard shows another, no stakeholder knows which to trust. The semantic layer resolves this by creating a single "golden layer"—a standardized interface that translates governed business language into physical database schemas, sitting between downstream consumers and raw tables.
Snowflake's internal adoption metrics illustrate the scale and velocity the semantic layer now supports: in a single month (July 2025), over 5,600 employees across sales, HR, support, and product teams ran more than 320,000 queries through the unified semantic layer. The company discovered concrete performance gains: in AtScale's benchmark testing, adding semantic context to text-to-SQL tasks increased accuracy from 20% to more than 90% on 40 business questions. For AI agents specifically, semantic views deliver two compounding returns—faster execution (agents avoid sampling and joining raw tables at query time) and lower token usage (pre-aggregated metrics and declared dimension-metric materializations reduce repeated computation). Snowflake's implementation treats the semantic layer as production software: views are version-controlled with dbt, undergo peer review and CI/CD, and are evaluated against real user questions logged in system logs. The company also leveraged Snowflake Cortex Agent's custom routing instructions and Cortex Sense to intelligently direct queries to the correct semantic view, and built automated materializations (pre-aggregated slices) that Snowflake maintains in the background rather than requiring manual pipeline management.
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytime
Ask AI anything about this article. Q&As are published on this page for other readers too.
Anthropic is privately hoping to file for its initial public offering by the end of this month, targeting a ra…
Broadcom is reportedly seeking to borrow up to $100 billion in debt financing to support growth efforts at Ant…
As AI technology matures, the bottleneck in the industry is moving beyond semiconductor constraints like GPUs…
Elice Group, a South Korean AI infrastructure provider, announced the launch of the country's first AI data ce…
On August 12, AT&T's Chief Data and AI Officer said OpenAI models power about 25% of the telecom's total AI us…
On August 11, IBM announced a multi-year $240 million agreement with Together AI to deploy NVIDIA HGX B300 sys…
