SnowPro Specialty: Gen AI Question 15
Single answerSnowflake CopilotA data engineering team has enabled Snowflake Copilot for analysts who frequently need help writing SQL against a curated sales schema. One analyst reports that Copilot can explain existing queries but often generates inaccurate SQL when asked to create new queries that join SALES_ORDERS, CUSTOMERS, and PRODUCTS. The tables exist in the same database, but the join logic is business-specific and not obvious from column names alone. The team wants to improve Copilot's usefulness without granting broader data access or moving data outside Snowflake. Which action is the best next step?
- A
Add clear object metadata such as comments and descriptive semantic context to the relevant tables and columns so Copilot has better schema understanding
- B
Increase the size of the virtual warehouse used by the analyst so Copilot can generate more accurate SQL joins
- C
Replicate the sales tables into a separate external vector database and connect Copilot to that system for schema reasoning
- D
Grant the analyst access to all schemas in the database so Copilot has more examples to infer the intended joins
Show answer and explanation
Correct answer: A
Explanation
The scenario points to a grounding and context problem, not a compute problem. Snowflake Copilot is most useful when the underlying Snowflake objects are well described and the user has appropriate access to the relevant data. When generated SQL is inaccurate because business-specific relationships are not obvious, improving object metadata such as table and column comments is the best first step. This aligns with Snowflake best practices around making data understandable and discoverable within Snowflake so AI-assisted tooling can generate better results. By contrast, increasing warehouse size does not improve semantic understanding, exporting metadata to external systems is unnecessary for Copilot in this scenario, and granting broader privileges conflicts with least-privilege access and the stated governance requirement.
- A. Correct.
Correct. Snowflake Copilot relies on available schema context and metadata to help users understand and generate SQL. If join relationships are business-specific and not obvious from names alone, adding meaningful comments and clear metadata to tables and columns is the most appropriate way to improve Copilot output while keeping data governance intact. This is a practical best practice because it improves the model's grounding in the objects the analyst is already authorized to use.
- B. Incorrect.
Incorrect. Warehouse size affects compute for SQL execution, not the language model's understanding of business semantics. Increasing warehouse size may improve query runtime after execution, but it does not address why Copilot is generating weak join logic in the first place.
- C. Incorrect.
Incorrect. Snowflake Copilot does not require exporting Snowflake schema information to an external vector database for this use case. Moving metadata or data outside Snowflake would add unnecessary complexity and could conflict with governance goals stated in the scenario.
- D. Incorrect.
Incorrect. Broader access is not the best next step because the requirement is to avoid granting additional data access. More schemas could also introduce more ambiguity rather than improving precision. Copilot responses are constrained by the user's privileges, and expanding those privileges is not a metadata-quality solution.