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
newsSNOWFLAKE

Snowflake turns document libraries into rows of AI summaries—but not structured facts

The AI_SUMMARIZE preview removes chunking and model selection from a common document workflow; teams still need extraction, validation and regional-limit controls.

Documents become summaries, but extracted facts stay separate.
Side by side: what changed
By The News Desk· Sep 15, 2026the quick take — two AI hosts go live when you do

Snowflake has put a multimodal summarizer behind a single SQL function. The September 14 public preview expands Cortex AI_SUMMARIZE to accept text or a staged file reference, including images and documents, and return a text summary. Snowflake says the managed function identifies themes, facts and relationships without a custom prompt, model choice or separate document-processing pipeline (release note).

The important part is the table boundary

The practical change is not merely that Snowflake can summarize one PDF. Its reference documentation shows users creating a table of file objects from a stage directory and then running SELECT AI_SUMMARIZE(doc) FROM doc_files to summarize every file as a row-producing SQL operation (function reference). That makes a document library composable with ordinary warehouse metadata: file name, content type, size, stage and modification time can be queried beside each generated summary (function reference).

This narrows an integration gap for data teams that already land contracts, reports or diagrams in Snowflake. They no longer need to select a model or build their own parsing and chunking pipeline for the summarization step, according to the release note (release note).

A summary is still not an extracted record

AI_SUMMARIZE returns text, not a typed set of fields. The docs also direct users to AI_AGG when summarizing across rows rather than concatenating rows before calling the function (function reference). Teams that need contract clauses, invoice amounts or other auditable fields should therefore treat this as a summarization stage, not a replacement for structured extraction and validation.

The operating boundaries matter. Inputs of 110 characters or fewer are returned unchanged, but processed tokens are still billed; row-level error details are optional; and Snowflake says multilingual quality can vary (function reference). File support also differs by region. The documented limits range from 4.5 MB and 100 pages for several document formats to 37.5 MB and 1,000 pages for PDF, TXT and Markdown in selected deployments, while Azure Global lists no document support in the current table (function reference).

For practitioners, the sensible rollout is a bounded one: start with one document class, retain the source-file metadata beside every summary, enable row-level error details, and evaluate factual retention before summaries feed search or agents. Snowflake has simplified the execution surface. It has not removed the need to test the output.

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