AI-Augmented SQL — Free Sample

1. An AI assistant produced a SQL query that joins `orders` to `customers` on `customer_id`. The query returns ~20% more rows than `SELECT COUNT(*) FROM orders`. What's the most likely cause and the right diagnostic next step?

The customers table has duplicate customer_id values; verify with `SELECT customer_id, COUNT(*) FROM customers GROUP BY 1 HAVING COUNT(*) > 1` before trusting the join output. The orders table has more rows than expected; recount `orders` directly to confirm. AI-generated queries always inflate row counts; rewrite the query without AI assistance. The join needs to be a LEFT JOIN; switching join type will fix it.

2. You're writing a query against a 50-table schema you don't fully know. What's the highest-leverage way to use AI assistance well?

Provide the AI with the relevant table schemas (column names, types, primary keys, foreign-key relationships) and the business question; verify the generated query's join paths against your understanding before running. Ask the AI to write the query from a high-level business description without schema context; trust the output if it runs without error. Ask the AI to write the query, run it on a small sample, and ship if the sample looks reasonable. Write the query yourself entirely; AI is unreliable for schemas you don't already know well.

3. An AI-generated query uses `GROUP BY 1, 2, 3` (positional grouping) on a complex aggregation. What's the right judgment call?

Replace positional grouping with explicit column names before committing; positional grouping is harder to read and breaks silently when SELECT-list ordering changes. Keep the positional grouping; it works correctly and is more concise. Reject the query; positional grouping is a code-smell that suggests AI hallucination. Run the query as-is; readability concerns are subjective.

4. You're debugging slow performance on a 100M-row analytical query. The AI assistant suggests adding `WHERE created_at >= CURRENT_DATE - INTERVAL '30 days'` to filter early. The suggestion improves runtime from 4 minutes to 8 seconds. What's worth verifying before shipping?

Confirm the business question actually wants only the last 30 days; the speedup is from filtering most of the data, which silently changes the analysis output. Trust the suggestion; AI optimization recommendations are reliable when they produce dramatic speedups. Add a comment explaining the optimization for future maintainers. Reject the change; AI shouldn't be making business-logic decisions like date-range filtering.

5. An AI assistant generates a query that uses a window function (`ROW_NUMBER() OVER (PARTITION BY ...)`) for deduplication. You've never used window functions on this warehouse before. What's the right next step?

Verify the warehouse supports the specific window-function syntax (vendors differ on edge cases like `QUALIFY` clauses), test on a small subset of partitions, confirm the dedup logic matches the business definition of duplicate. Ship it; window functions are standard SQL and behave consistently across warehouses. Reject and rewrite without window functions; they're harder for the team to maintain. Run as-is and rollback if the output looks wrong.

Your directional indicator

This 5-question sample gives a rough indication only. For a valid, verified Skills Passport credential, take the full 40-question assessment — it takes about 30 minutes.

Coming soon

Full assessment launching soon