Snowflake CoWork+36 Positive

GILD Review

GILD's call and the buying facts behind it.

From GILD research

Should we adopt Snowflake CoWork?

Will your business users trust Snowflake CoWork's answers enough to stop opening their dashboards?

30-Second TakePick Snowflake CoWork if your data team maintains semantic views and can test them before each team gets access.
Best fitSnowflake customers with a modeled warehouse and a data team ready to own semantic views and agents.GILD Review
Consider another workflowIf your data spans several platforms, a workspace such as Hex reaches more of it; if your metrics already live in LookML, evaluate Looker's Conversational Analytics first.GILD Review

Drawn from 77 sources and 2 operator insights

GILD analysis

Snowflake CoWork Review

Buy CoWork if your data team already maintains Snowflake semantic views and has time to test them before each new team gets access. It answers under each user's own roles, and every question draws AI Credits plus warehouse compute. Accuracy rides on semantic views and verified queries that the data team keeps current and tests with evaluations.

AI Data Analysis
AI Adoption

The decision

Best fit: Snowflake customers with a modeled warehouse and a data team ready to own semantic views and agents.
Consider another workflow: If your data spans several platforms, a workspace such as Hex reaches more of it; if your metrics already live in LookML, evaluate Looker's Conversational Analytics first.
Adoption condition: A semantic view with verified queries, a passing evaluation set, and a resource budget for each agent before it opens to a team.

Basis: Snowflake documentation and pricing, plus public accounts from teams that used it, including Mastech's 30-question accuracy evaluation of Cortex Analyst. Snowflake reported in August 2024 that GPT-4o alone answered 51% of its 150 real-world BI questions correctly, so score your own evaluation set before launch.

Buying & deployment factsWhat business and product leaders should verify before rolling it out across teams.

How you buy it

US$2.00

Per AI Credit on global routing, with no seat fee. Regional routing is US$2.20.

Source 1, Snowflake, AI pricing: AI pricing Source 2, Snowflake, Service Consumption Table: Service Consumption Table Source 3, Snowflake, User access and settings: User access and settings

Details

CoWork has no per-seat fee and runs on an existing Snowflake account. It bills AI Credits, separate from Platform Credits, at US$2.00 per AI Credit on global routing and US$2.20 on regional routing, on demand. Contract-value discounts bring them down to US$1.88 and US$2.07. AI Credit prices are the same on every edition.

Cost drivers

Tokens plus tools

Each model's token rate, plus every tool the agent calls and the warehouse running its SQL.

Source 1, Snowflake, AI pricing: AI pricing Source 2, Snowflake, Service Consumption Table: Service Consumption Table

Details

Token use and tools drive cost. Orchestration bills per million tokens at the selected model's rate. For example, claude-sonnet-5 costs 1.30 AI Credits per million input tokens and 6.50 per million output tokens. Costs add across the tools an agent calls, such as Cortex Search at 6.3 AI Credits per GB of indexed data per month. The generated SQL runs on a warehouse billed in Platform Credits.

Where it runs

Your Snowflake account

At ai.snowflake.com, in Snowsight, and on iOS, with each user's own credentials.

Source 3, Snowflake, User access and settings: User access and settings Source 4, Snowflake, CoWork overview: CoWork overview Source 5, Snowflake, Cross-region inference: Cross-region inference

Details

In the customer's Snowflake account, used at ai.snowflake.com, inside Snowsight, and in an iOS app, with AWS PrivateLink and Azure Private Link supported. Queries run with each user's own credentials, and frontier models need cross-region inference.

What to verify

CORTEX_USER

Granted to PUBLIC by default. Revoke it and grant agent access to chosen users.

Source 3, Snowflake, User access and settings: User access and settings Source 5, Snowflake, Cross-region inference: Cross-region inference Source 6, Snowflake, Cortex AI Guardrails: Cortex AI Guardrails Source 4, Snowflake, CoWork overview: CoWork overview

Details

CORTEX_USER, which opens every Cortex feature, is granted to PUBLIC by default. Revoking it and granting CORTEX_AGENT_USER limits chosen users to agents and CoWork. Model availability depends on region and cross-region policy, and Cortex AI Guardrails require Enterprise Edition. Snowflake classifies CoWork inputs as Usage Data and outputs as Customer Data.