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
newsGOOGLE CLOUD

Google’s Data Agent Kit reaches GA—but the agent still spends your cloud budget

The free toolkit can inspect and change more than 15 Google data services; its real production boundary is inherited IAM plus metered queries and jobs.

Agent-assisted data work versus the inherited cloud controls and costs it must obey.
AI-generated illustration
By The News Desk· Oct 1, 2026the quick take — two AI hosts go live when you do

Google Cloud has made Data Agent Kit generally available, turning a preview developer aid into a supported bridge between coding agents and more than 15 Google Data Cloud services. The kit combines Model Context Protocol tools with Google-authored skills and works from VS Code-compatible editors, Cloud Shell, Cloud Workstations, Claude Code, Codex and Google’s Antigravity tools. Google says the kit itself has no additional charge, while the cloud services invoked by an agent retain their standard pricing. Source: Google Cloud’s GA announcement

GA expands the execution surface

The GA release adds BigQuery Graph, Bigtable and Managed Service for Apache Spark. An agent can map relational tables into a proposed property graph, generate GQL, design a Bigtable row key, query Iceberg tables through serverless Spark, or connect an open lakehouse to AWS Glue and Databricks Unity Catalog. It can also turn working logic into dbt or Dataform models and schedule those with notebooks through Managed Service for Apache Airflow. Source: Google Cloud’s GA announcement

That breadth is the important change for data teams: the same conversational interface can move from schema discovery and SQL generation into resource creation, pipeline orchestration and troubleshooting. Google says graph creation includes a plan for the user to approve, and failed Airflow or Spark tasks can be handed to the agent for a proposed fix. Those approval points matter because this is no longer only a query-writing assistant. Source: Google Cloud’s GA announcement

Identity is the control plane

Data Agent Kit acts as the signed-in user or an impersonated service account. Google says existing row- and column-level policies therefore apply, while administrators can govern MCP access through IAM, inspect traffic with Model Armor, and constrain agents with VPC Service Controls and Principal Access Boundary policies. In practice, teams should treat the kit as another execution client: an overly broad human or service-account role becomes an overly broad agent role. Source: Google Cloud’s GA announcement

Cost is a separate boundary. The toolkit is free, but BigQuery scans, Spark sessions, database operations and other invoked services are not. Google says its skills encourage cost-aware patterns such as checking partition keys and running BigQuery dry runs before execution. That is useful guidance, not a hard budget ceiling; production rollouts still need project-level quotas, billing alerts and narrowly scoped identities around agent workloads. Source: Google Cloud’s GA announcement

The practical deployment test is therefore not whether the agent can write SQL. It is whether every tool call inherits the right identity, presents a review step before consequential changes, and stays inside a measurable spend envelope. GA makes the integration easier; it does not transfer ownership of those controls from the platform team to the agent.

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