AI coding: SQL by explanation-first, query-second

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-ai-coding-sql-query-explain-then-write-r7a1-creators

1. What is the most common silent failure mode of AI-generated SQL?

   1. Using lowercase keywords
   2. Joining the wrong way or aggregating over the wrong grain
   3. Adding too many comments
   4. Inserting random emojis

2. Why ask the AI to explain in English before producing SQL?

   1. So you can catch the misunderstanding before running anything against production
   2. Because English runs faster than SQL
   3. To avoid using indexes
   4. Because AI cannot write SQL otherwise

3. Which item should the English plan name explicitly?

   1. The keyboard layout
   2. Tables, join keys, grain, and aggregation
   3. The database vendor's CEO
   4. The DBA's lunch order

4. Which task is AI well suited to once given a schema?

   1. Knowing your warehouse's actual cardinality
   2. Translating an English question into syntactically valid SQL
   3. Setting cost budgets for queries
   4. Choosing which BI tool you should buy

5. Which SQL feature can AI use confidently when you ask for it?

   1. Window functions and CTEs
   2. Telepathy with your DBA
   3. Predicting last week's row counts exactly
   4. Reading uncommitted transactions on other systems

6. What hidden bug does a missing join key often produce?

   1. A syntax error at compile time
   2. A silent cartesian fan-out that explodes at production scale
   3. A network outage
   4. A schema migration

7. Which check guards against silent fan-out?

   1. Compare row counts against an expected upper bound
   2. Run the query twice
   3. Use a different IDE
   4. Rename the columns

8. Which limitation should you assume when asking AI for SQL?

   1. It knows your schema's cardinality and skew
   2. It cannot decide acceptable query cost on your warehouse
   3. It will always add the cheapest indexes automatically
   4. It can read your production data

9. Which prompt opening best fits explanation-first SQL?

   1. 'Just give me the query.'
   2. 'First, explain in plain English: tables, join keys, grain, aggregation. Wait for me to confirm before writing SQL.'
   3. 'Use as many subqueries as possible.'
   4. 'Make it a one-liner.'

10. At which scale do silent join bugs most often surface?

    1. The first three rows in your dev database
    2. Production data with realistic cardinality
    3. Static example queries in documentation
    4. Hand-typed test rows

11. What is 'verifiable output' for AI-generated SQL?

    1. Output you can independently check against a known truth or bound
    2. Output that passes a syntax linter only
    3. Output formatted in lowercase
    4. Output longer than 50 lines

12. Why does AI need indexes suggested only after seeing the query plan?

    1. Because the plan exposes which scans and joins actually dominate
    2. Because plans are decorative
    3. Because indexes are forbidden by SQL
    4. Because indexes always slow down a database

13. Which mistake does explanation-first SQL specifically protect against?

    1. Color theme drift
    2. Aggregating revenue at the wrong grain (e.g., per row instead of per order)
    3. CSS misalignment
    4. Browser cache invalidation

14. Which workflow earns the most trust in AI-written SQL?

    1. Run it directly on production immediately
    2. Confirm the English plan, run on a sample, sanity-check counts, then promote
    3. Email the query to the team without testing
    4. Paste it into a slide deck

15. What is the upside of AI for SQL even after these caveats?

    1. It removes the need for any human review
    2. It accelerates drafting and surfaces options you can verify
    3. It guarantees optimal queries every time
    4. It eliminates the need for a database