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
guideEnterprise guide

Snowflake’s marketplace agent can recommend a product—but it cannot outrun your privileges

CoCo now carries discovery through comparison, trial and installation. The important deployment detail is that conversation changes the interface, not the authorization model.

Chat can recommend and start actions, but Snowflake permissions still gate installation.
AI-generated illustration
By The News Desk· Oct 15, 2026the quick take — two AI hosts go live when you do

Snowflake has put a conversational agent in front of its public Marketplace, turning product discovery into a workflow that can continue through comparison, trial and installation. The public-preview feature, called Agentic Marketplace Discovery, lets a user describe a need to CoCo, inspect matching listings and act on one without leaving the conversation. [Snowflake release note]

The practical boundary is easy to miss: CoCo changes how a user reaches an action, not what the user is authorized to do. Snowflake says the feature is available to all accounts, but using CoCo for Marketplace work requires the SNOWFLAKE.COPILOT_USER database role plus either SNOWFLAKE.CORTEX_USER or SNOWFLAKE.CORTEX_AGENT_USER. Installing a listing or starting a trial still requires the privileges for the underlying consumer action, including privileges such as CREATE DATABASE and IMPORT SHARE. [Snowflake documentation]

The chat can cross the purchase funnel

Agentic discovery can search in natural language, recommend a fit, compare coverage, granularity or refresh cadence, and explain how a listing might enrich data already in the account. From the same chat, a user can install an accessible listing, start an offered trial, open a sales request or initiate a request for a private listing. Listing cards can identify commercial and delivery attributes such as Free to try, Paid, Secure share or Native App. [Snowflake documentation]

That makes the feature more consequential than a conversational search box. It joins product selection and acquisition in one interface. For platform teams, the control point therefore belongs at the roles that permit acquisition—not only at access to CoCo.

Snowflake preserves several explicit checkpoints. CoCo uses the same consumer agreements and access rules as the Marketplace interface. Before an install, trial or request proceeds, the user must complete the required acknowledgments and configuration; an installation form can also ask the user to review the database name. Billing, payment setup and capacity-drawdown rules still apply to paid listings. [Snowflake documentation]

What administrators should test

Administrators evaluating the preview should separate three permissions in their test plan: who can converse with CoCo, who can discover public listings, and who can create the database or import the share behind an installation. A successful recommendation does not prove that the resulting action is authorized, and a blocked installation may be the intended role policy rather than an agent failure.

The preview also has a firm scope limit: it works only with the public Snowflake Marketplace, not the internal marketplace. It therefore does not replace internal catalog or procurement workflows. [Snowflake documentation]

The useful mental model is an agent-assisted storefront with the existing checkout controls intact. Teams should evaluate the relevance of recommendations, but they should audit the action boundary separately—especially wherever Marketplace privileges are held by broad analyst roles.

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