live wire
▸Snowflake agent evals can pin judge versions; unpinned metrics may roll forward and shift scoresSnowflake▸ThoughtSpot makes Spotter 3 default for new instances; it requires download rights and always shares query data with an LLMThoughtSpot documentation▸MotherDuck adds image classification and object location to prompt_jev(), returning typed SQL outputs with confidence scoresMotherDuck▸BigQuery conversational analytics adds Montréal and Toronto; mixed-region agents still default to globalGoogle Cloud Documentation▸Fabric enables Azure OpenAI Copilot access by default, but OpenAI-subprocessor and cross-region processing/storage switches remain offMicrosoft Learn▸Snowflake Cortex Agents now continue by default when named tools are inaccessible, emitting warnings instead of failing the runSnowflake documentation▸Databricks Genie One can use group memberships to tailor answers, while Unity Catalog remains the access-control boundaryDatabricks documentation▸Fabric data agents can query Power BI models with Read alone; Build and workspace access are unnecessaryMicrosoft Learn▸Fabric data-agent Git configs expose column flags, but selecting a table still includes every columnMicrosoft Learn▸Google Data Agent Kit offers remote Spark kernels, but notebook exports can still move results into local Pandas or ArrowGoogle Cloud Documentation▸Cube says production analytics agents need explicit partial-result states, caller-filtered model search and durable artifactsCube▸Thinking Machines names Bridgewater behind ReViSQL-K2.6, citing 92.97% SC-16 accuracy at 15% of frontier-model costThinking Machines Lab▸AWS says a mismatched data/tool/prompt release cut its Field Sales agent pass rate from 78% to 38% in pre-productionAWS Artificial Intelligence Blog▸NL2SQL Multi-Agent ships evaluation artifacts across BIRD, CSpider, Archer and KDD DataSpace—but no software licenseGitHub▸Snowflake agent evals can pin judge versions; unpinned metrics may roll forward and shift scoresSnowflake▸ThoughtSpot makes Spotter 3 default for new instances; it requires download rights and always shares query data with an LLMThoughtSpot documentation▸MotherDuck adds image classification and object location to prompt_jev(), returning typed SQL outputs with confidence scoresMotherDuck▸BigQuery conversational analytics adds Montréal and Toronto; mixed-region agents still default to globalGoogle Cloud Documentation▸Fabric enables Azure OpenAI Copilot access by default, but OpenAI-subprocessor and cross-region processing/storage switches remain offMicrosoft Learn▸Snowflake Cortex Agents now continue by default when named tools are inaccessible, emitting warnings instead of failing the runSnowflake documentation▸Databricks Genie One can use group memberships to tailor answers, while Unity Catalog remains the access-control boundaryDatabricks documentation▸Fabric data agents can query Power BI models with Read alone; Build and workspace access are unnecessaryMicrosoft Learn▸Fabric data-agent Git configs expose column flags, but selecting a table still includes every columnMicrosoft Learn▸Google Data Agent Kit offers remote Spark kernels, but notebook exports can still move results into local Pandas or ArrowGoogle Cloud Documentation▸Cube says production analytics agents need explicit partial-result states, caller-filtered model search and durable artifactsCube▸Thinking Machines names Bridgewater behind ReViSQL-K2.6, citing 92.97% SC-16 accuracy at 15% of frontier-model costThinking Machines Lab▸AWS says a mismatched data/tool/prompt release cut its Field Sales agent pass rate from 78% to 38% in pre-productionAWS Artificial Intelligence Blog▸NL2SQL Multi-Agent ships evaluation artifacts across BIRD, CSpider, Archer and KDD DataSpace—but no software licenseGitHub
nl2sql.ai
analysisANALYSIS

Google’s new Gemini lakehouse path crosses clouds—but catalog access is not data relocation

A preview connects natural-language questions to BigQuery and Databricks Unity Catalog through federation, while Google’s architecture keeps execution, metadata and network paths distinct.

Gemini fronts federated catalogs while data stays in place.
Side by side: what changed
By The News Desk· Oct 15, 2026the quick take — two AI hosts go live when you do

Google Cloud has previewed a Gemini integration for its borderless Lakehouse that lets business users query records across BigQuery and Databricks Unity Catalog in natural language. The important architectural point is what Google says does not happen: the underlying data does not have to be moved into Gemini or copied into a single warehouse first.

Gemini becomes the front door, not the query engine

The preview allows teams to configure access to sources including Databricks Unity Catalog and Snowflake Horizon through Lakehouse catalog federation. Gemini can then ground agent reasoning and generate code against that distributed estate. In Google’s description, the agent is the conversational entry point; the lakehouse, catalogs and query engines still do the large-scale work.

That distinction matters because an agent context window is not a substitute for a database execution layer. Google explicitly contrasts the hundreds of kilobytes that an agent may ingest with enterprise joins spanning petabytes, and says MCP payloads remain constrained to megabyte scale. The design therefore pushes computation to the data systems and returns only the context needed by the agent.

“Without moving data” still has infrastructure costs

Federation avoids a migration project, but it does not make cross-cloud reads free or instantaneous. Google says every uncached byte scanned across a cloud boundary still traverses the network. Its Cross-Cloud Interconnect offers private 10 Gbps or 100 Gbps connections to AWS or Azure, while cross-cloud caching serves repeated reads from Google Cloud. Operators should read the launch as a choice of execution and caching architecture—not as a claim that distance has disappeared.

Several supporting capabilities are now generally available: cross-cloud caching, Apache Iceberg V3 support, read-write Iceberg interoperability for BigQuery and Managed Service for Apache Spark, AlloyDB lakehouse federation, and bidirectional catalog caching with AWS Glue, Unity Catalog and Snowflake Horizon. The Gemini access itself remains preview.

Metadata is the control plane

Google positions the Lakehouse runtime catalog and Knowledge Catalog as the layer that gives agents business definitions and asset metadata without stuffing entire datasets into prompts. The runtime catalog uses the Iceberg REST Catalog model and adds federation plus a Hive Metastore-compatible endpoint. Knowledge Catalog indexes distributed assets, including unstructured objects in Cloud Storage.

For practitioners, the practical boundary is clear: Gemini can unify the question interface, but correctness and governance still depend on catalog synchronization, semantic metadata, source permissions, caching policy and the execution engine behind each request. A single chat surface does not collapse those responsibilities; it makes their coordination the product.

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