live wire
IBM makes watsonx Orchestrate AgentOps, custom LLM judging and Bedrock-agent discovery generally availableIBMSchemaGate 0.1.45 fixes broken Oracle ADB wallet connections and an OCI stack pinned 28 releases behindSchemaGatePDI’s Amazon Quick procurement agent grounds spend answers in vendor, category and contract contextAWS Business Intelligence BlogBigQuery’s ML.METRICS example returns 0.84 accuracy but 0.30 macro-F1 on the same 100-row classification queryGoogle Cloud BigQuery docsSchemaGate 0.1.44 auto-selects sentence embeddings, lifting bundled-schema retrieval from 90/98 to 93/98SchemaGateSchemaGate 0.1.43 adds read-only SQL execution with per-principal table checks—and documents unauthenticated client assertionsSchemaGateDatabox adds reusable AI Analyst Skills with personal/company scope, auto-matching and marketplace installsDataboxFabric previews an AI builder for data-agent instructions, source guidance and example queriesMicrosoft FabricDatabricks trains data-agent retriever to stop early or spend bounded extra search steps, reporting 5.8-second latencyDatabricksThoughtSpot adds SpotterCode coding agent to its Visual Embed PlaygroundThoughtSpotLongMemEval-S audit: 67–73% of restore-fixable 80k-budget errors came from evicted evidence under three policiesarXivSchemaGate 0.1.42 adds dimension-aware retrieval and fixes complex multi-table SQL promptsSchemaGateSnowflake agent toolsets can silently drop inherited tools when callers lack accessSnowflake DocumentationLooker’s VS Code extension reaches GA with MCP-assisted LookML generation, editing and validationGoogle Cloud Looker release docsIBM makes watsonx Orchestrate AgentOps, custom LLM judging and Bedrock-agent discovery generally availableIBMSchemaGate 0.1.45 fixes broken Oracle ADB wallet connections and an OCI stack pinned 28 releases behindSchemaGatePDI’s Amazon Quick procurement agent grounds spend answers in vendor, category and contract contextAWS Business Intelligence BlogBigQuery’s ML.METRICS example returns 0.84 accuracy but 0.30 macro-F1 on the same 100-row classification queryGoogle Cloud BigQuery docsSchemaGate 0.1.44 auto-selects sentence embeddings, lifting bundled-schema retrieval from 90/98 to 93/98SchemaGateSchemaGate 0.1.43 adds read-only SQL execution with per-principal table checks—and documents unauthenticated client assertionsSchemaGateDatabox adds reusable AI Analyst Skills with personal/company scope, auto-matching and marketplace installsDataboxFabric previews an AI builder for data-agent instructions, source guidance and example queriesMicrosoft FabricDatabricks trains data-agent retriever to stop early or spend bounded extra search steps, reporting 5.8-second latencyDatabricksThoughtSpot adds SpotterCode coding agent to its Visual Embed PlaygroundThoughtSpotLongMemEval-S audit: 67–73% of restore-fixable 80k-budget errors came from evicted evidence under three policiesarXivSchemaGate 0.1.42 adds dimension-aware retrieval and fixes complex multi-table SQL promptsSchemaGateSnowflake agent toolsets can silently drop inherited tools when callers lack accessSnowflake DocumentationLooker’s VS Code extension reaches GA with MCP-assisted LookML generation, editing and validationGoogle Cloud Looker release docs
nl2sql.ai
guideImplementation guide

Oracle’s conversational export pattern keeps the agent outside the trust boundary

A new Oracle tutorial turns a Data Transforms export into an MCP tool—but the useful part is the narrow wrapper, duplicated confirmation, and honest asynchronous status.

Narrow wrapper versus broad access around a database export path.
AI-generated illustration
By The News Desk· Sep 9, 2026the quick take — two AI hosts go live when you do

Oracle has published a tested pattern for triggering an Oracle Data Transforms export from natural language without giving the AI client direct SQL, credentials, or arbitrary network access. The September 8 tutorial wraps Data Transforms’ existing REST export flow in PL/SQL, registers a narrow function as a Select AI Agent tool, and exposes it through the Autonomous AI Database MCP Server.

The database stays in charge

The architecture separates duties across two schemas. An export-owner schema holds the procedure that authenticates to Data Transforms and calls its REST API. A separate MCP login schema exposes only a wrapper function and receives an explicit cross-schema EXECUTE grant. Oracle says the client therefore sees one approved operation rather than the underlying credentials, unrestricted PL/SQL, or a general-purpose HTTP capability.

That distinction matters more than the conversational interface. The model chooses a named tool; the database decides what that tool can actually do. Data Transforms still creates the export job and writes the ZIP archive to the configured Object Storage connection. MCP changes the invocation surface, not the system of record or execution path.

Confirmation exists at two layers

The wrapper refuses to proceed unless its input equals the literal string EXPORT. The registered tool instruction separately tells the client to call the tool only after an explicit request and confirmation, and no more than once for a single request. Oracle describes this as a belt-and-suspenders design: even if the model’s tool-selection behavior is imperfect, the function itself still enforces a deterministic gate.

The tool definition also says returned output must be treated as data rather than instructions. That is a practical prompt-injection defense because the REST response is echoed back to the client largely unfiltered. Outbound access is constrained too: Autonomous AI Database blocks the HTTP call by default, so the executing schema needs an explicit network ACL grant for the Data Transforms host.

“Submitted” is not “finished”

The sharpest operational detail is the status boundary. A successful direct database test proves DNS, TLS, authentication, and export submission; it does not prove the asynchronous job completed. The example response reports the job as Submitted (WAITING), and Oracle says production use should add job-status monitoring, duplicate-request protection, archive verification, audit logging, and secret management.

That makes this less a story about replacing console clicks and more a reusable deployment rule: expose side effects as small, typed operations; enforce approval below the model; and report workflow state precisely. The agent can open the door, but the database keeps the keys—and operators still need to verify what happened after the call returned.

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.