live wire
▸Oracle Select AI generates SQL/PGQ for property graphs—but errors if one AI profile mixes graphs with tables or viewsOracle▸Domo’s data team standardizes MCP builds with shared scaffolding; beta toolkits inherit user permissionsDomo▸Databricks proposes five outcome signals for agentic BI—and says dashboard counts and license usage miss decision qualityDatabricks▸One vibe-coded database app needed 56+ human test hours and three repair rounds despite AI Studio’s verification claimsMancas & Mancas / arXiv▸BigQuery AI.IF’s default cost mode trains a local model for most rows—and drops row-error thresholdsGoogle Cloud▸AWS’s multi-tenant data-agent pattern gives the broker Lambda no data permissions; Cognito roles carry Lake Formation scopeAWS Big Data Blog▸Amazon OpenSearch’s optimized engine adds nested agent traces, but AWS reserves it for log analytics—not vector searchAWS▸Two interpretation-based provenance semantics cannot express SQL bag difference, new theorem showsKolaitis / arXiv▸Databricks Feature Views reuse one definition for training and live scoring, but online lookup still returns latest available stateDatabricks▸Databricks Genie’s Slack and Teams previews default channel answers to public, even when viewers lack source-data accessDatabricks▸Snowflake’s two-tool GraphRAG agent beat a seven-tool graph setup 70% to 60% on 22 ontology questionsSnowflake Engineering▸Semantix opens an Apache-2.0 semantic-first data-agent stack, but labels the project WIP and has no release or adoption yetGitHub / icosmos-space▸Confluent defines Streamhouse as shared current state for agents and apps, but says broader semantics and governance are still to comeConfluent▸Graph-repair retrieval cut F1 by 0.126 on one suite; a learned router used fewer tokens than always-on retrieval in all sixLin et al. / arXiv▸Oracle Select AI generates SQL/PGQ for property graphs—but errors if one AI profile mixes graphs with tables or viewsOracle▸Domo’s data team standardizes MCP builds with shared scaffolding; beta toolkits inherit user permissionsDomo▸Databricks proposes five outcome signals for agentic BI—and says dashboard counts and license usage miss decision qualityDatabricks▸One vibe-coded database app needed 56+ human test hours and three repair rounds despite AI Studio’s verification claimsMancas & Mancas / arXiv▸BigQuery AI.IF’s default cost mode trains a local model for most rows—and drops row-error thresholdsGoogle Cloud▸AWS’s multi-tenant data-agent pattern gives the broker Lambda no data permissions; Cognito roles carry Lake Formation scopeAWS Big Data Blog▸Amazon OpenSearch’s optimized engine adds nested agent traces, but AWS reserves it for log analytics—not vector searchAWS▸Two interpretation-based provenance semantics cannot express SQL bag difference, new theorem showsKolaitis / arXiv▸Databricks Feature Views reuse one definition for training and live scoring, but online lookup still returns latest available stateDatabricks▸Databricks Genie’s Slack and Teams previews default channel answers to public, even when viewers lack source-data accessDatabricks▸Snowflake’s two-tool GraphRAG agent beat a seven-tool graph setup 70% to 60% on 22 ontology questionsSnowflake Engineering▸Semantix opens an Apache-2.0 semantic-first data-agent stack, but labels the project WIP and has no release or adoption yetGitHub / icosmos-space▸Confluent defines Streamhouse as shared current state for agents and apps, but says broader semantics and governance are still to comeConfluent▸Graph-repair retrieval cut F1 by 0.126 on one suite; a learned router used fewer tokens than always-on retrieval in all sixLin et al. / arXiv
nl2sql.ai
guideCONTEXT ENGINEERING

Verified SQL examples are not portable across Microsoft, Google and Snowflake data agents

The same question-and-query pair can act as a retrieved hint, a composable blueprint or a governed semantic-model asset. Treating those roles as interchangeable is a migration trap.

By The News Desk· Nov 7, 2026the quick take — two AI hosts go live when you do

A team moving an NL2SQL workload between enterprise platforms can carry over its question-and-SQL pairs and still change the system’s behavior. Microsoft, Google and Snowflake all let operators provide curated examples, but their documentation assigns those examples materially different jobs.

Microsoft: examples are retrieved hints

In Microsoft Fabric Data Agent, SQL and KQL sources accept natural-language/query pairs. The service retrieves the most relevant examples with vector similarity, then uses them alongside selected schema and instructions. Microsoft separately validates generated SQL and KQL against the schema entities that the operator approved.

That makes an example part of retrieval-time guidance, not the whole authorization boundary. Fabric also handles Power BI semantic models differently: instructions and Verified Answers are managed through Power BI’s Prep for AI surface rather than as ordinary data-source instructions. One product therefore already has multiple example contracts depending on the source type.

Google: blueprints are structures that facets can modify

Google’s preview QueryData service puts validated question/SQL pairs inside a broader context set. Its query blueprints define canonical intent-to-SQL mappings, but query facets contribute reusable predicates and value searches map informal phrases to actual column values. QueryData can also return a clarification question when an input is ambiguous.

A Google blueprint is therefore not merely a nearest-neighbor demonstration. It is one component in a synthesis process that can resolve values, combine filters and ask for disambiguation before generating SQL. Migrating only the paired examples would omit two other documented context primitives.

Snowflake: verification is a governed semantic-model asset

Snowflake’s Cortex Analyst Verified Query Repository stores questions and SQL inside a semantic model or view. The SQL must reference logical table and column names from that model rather than the underlying physical names. Snowflake’s documented workflow asks a person to inspect, edit, execute and test the generated query before saving it as verified; optional fields record who verified it and when.

Cortex Analyst can reuse a relevant verified query for a similar question, and the API confidence information can show which verified query contributed to a response. The repository is thus both a grounding mechanism and a provenance-bearing review artifact.

The migration test

Before transferring an example library, ask four questions: How is an example selected? Can other context rewrite or compose it? Which names does its SQL bind to? What review and provenance survive deployment?

A high-quality pair can be useful on all three platforms, but it is not the same control object. Microsoft treats examples as retrieved guidance within a source configuration; Google makes blueprints composable with facets and value resolution; Snowflake binds verified SQL to logical semantic-model names and a human review trail. Port the intent and tests—not just the YAML or JSON rows.

sources

Filed by The News Desk. Corrections: desk@nl2sql.ai · Our standards →

comments · 0

    Comments are moderated before they appear. Your email is used once to confirm it is you — never shown, never sold. Corrections and questions get an answer from the desk when we have one.

    more from the desk