SnowPro Specialty: Gen AI Question 16
Single answerSnowflake CopilotA data engineering team enables Snowflake Copilot for analysts who frequently explore unfamiliar schemas. One analyst opens Snowsight, selects a database and schema, and asks Copilot: "Show total monthly revenue by region for the last 12 months." Copilot generates SQL, but the analyst notices the result uses the ORDERS table instead of the FINANCE_REVENUE table that contains the governed revenue metric the company requires. The team wants analysts to keep using Copilot while reducing the chance of Copilot selecting the wrong objects or business logic. What is the best action to take?
- A
Create or expose a semantic model layer with trusted business definitions so Copilot can ground its SQL generation on approved metrics and relevant objects.
- B
Increase the size of the virtual warehouse used by analysts so Copilot has more compute available to choose the correct table and metric definitions.
- C
Grant analysts OWNERSHIP on all tables in the database so Copilot can inspect every object and infer the right revenue logic automatically.
- D
Disable role-based access controls for the FINANCE schema so Copilot can reference protected objects even when the analyst's role does not have access.
Show answer and explanation
Correct answer: A
Explanation
This scenario is about improving the quality and governance of Copilot-generated SQL, not query performance or bypassing security. The best practice is to provide Copilot with curated business context through trusted semantic definitions, governed metrics, and relevant objects so that natural language requests resolve to the intended tables and calculations. In Snowflake, generative AI assistants such as Copilot work best when grounded in metadata and organizational context, while still respecting role-based access controls. Increasing warehouse size does not affect semantic accuracy, and granting excessive privileges or disabling RBAC violates core Snowflake security principles. Candidates should recognize that for business-critical metrics like revenue, the right fix is to improve the governed semantic layer and context available to Copilot rather than changing compute or weakening security.
- A. Correct.
Correct. Snowflake Copilot is more reliable when grounded in trusted metadata and business context rather than only raw physical tables. Using a semantic model or curated layer with approved metric definitions helps Copilot generate SQL against the intended objects and business logic, which is exactly the issue in this scenario.
- B. Incorrect.
Incorrect. Warehouse size affects query execution resources, not Copilot's understanding of business semantics or object selection. More compute does not improve the governance of generated SQL or make Copilot choose the correct revenue table.
- C. Incorrect.
Incorrect. Broadening privileges is not a best practice and does not solve the core problem of ambiguous business definitions. It may even increase the chance of Copilot choosing from too many similar objects. Snowflake access should follow least privilege.
- D. Incorrect.
Incorrect. Copilot operates within Snowflake's security model; disabling access controls is not an appropriate or supported governance approach. Analysts should only be able to access objects permitted by their role, and improving grounding through trusted semantic definitions is the right solution.