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
newsMajor launch

OpenAI turns ChatGPT Work into a governed analytics front end

The new Data agent connects across major warehouses and BI tools while inheriting existing permissions—a direct bid to make ChatGPT the interface above the enterprise data stack.

ChatGPT Work as governed analytics front end over existing data stack
AI-generated illustration
By The News Desk· Sep 10, 2026the quick take — two AI hosts go live when you do

OpenAI has launched a Data agent inside ChatGPT Work that can investigate company data, refine an analysis conversationally and build refreshable interactive dashboards. The larger move is architectural: OpenAI is positioning ChatGPT as a common analytics interface above warehouses, semantic layers and existing BI systems rather than asking companies to replace them.

According to OpenAI’s launch post, approved connections include Amazon Redshift, Google BigQuery, ClickHouse, Databricks, MongoDB and Snowflake, alongside files from Google Drive and SharePoint. The agent can also build and interact with dashboards in Omni, Oracle BI, Power BI, Sigma, Tableau and ThoughtSpot.

Context and permissions stay underneath

The pitch is not raw schema-to-SQL generation. OpenAI says the agent consumes business terms, metric definitions, custom calculations and data relationships from sources including Databricks Genie Ontology, dbt, GitHub, Snowflake Horizon and BI dashboards. Workspace administrators choose which connections and roles are available, while queries inherit the connected account’s table-, row- and column-level restrictions.

That makes the product a coordination layer across an existing stack. The warehouse still executes queries. Semantic and catalog systems still define meaning. BI tools can remain presentation targets. ChatGPT owns the conversation, investigation flow and a new dashboard surface.

The security boundary, however, depends on each connector. ClickHouse’s launch implementation combines its managed Remote MCP server with Apache-2.0 agent skills. OAuth limits access to organizations and services visible to the ClickHouse Cloud user, while 13 exposed tools can inspect schemas, execute SELECT statements and read operational metadata. ClickHouse says every tool is read-only and declares readOnlyHint: true; vendor-maintained skills provide guidance on schema design, querying and ingestion without expanding the MCP server’s permissions.

From answers to approved actions

OpenAI says users can review evidence behind findings, turn analyses into shareable dashboards, and ask the agent to recommend next steps. It can distribute findings through Slack or email and carry out actions through connected tools when the user approves them. That extends the product beyond NL2SQL, but it also makes the approval and connector boundaries more important than the fluency of the generated narrative.

OpenAI reports that nearly all of its product staff and more than two-thirds of its go-to-market organization use the internal capabilities behind the Data agent. It also names NTT Data, Thermo Fisher and ServicePiston among alpha users. Those are adoption claims from OpenAI, not independently reported performance measurements; the launch post does not publish accuracy, latency or cost benchmarks.

For data teams, the immediate competitive question is no longer only which assistant writes the best SQL. It is which layer becomes the place where employees start an investigation. OpenAI is betting that a cross-platform conversation—with governed connectors and dashboards attached—can become that place while the existing data stack remains underneath.

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