Where AI helps with your database, and where it goes blind

AI is a strong drafting partner for database work and a poor substitute for your own knowledge of the data. Knowing the split keeps you fast and safe.

Where AI helps:

• Drafting queries from a plain description once you supply the schema.
• Explaining unfamiliar SQL, stored procedures, or a dense query someone left behind.
• Suggesting schema shapes, index candidates, and migration steps to consider.
• Turning an error message into a plausible cause and a fix to check.

Where AI goes blind:

• It cannot see your real rows, their volumes, or their skew. A query that looks fine can scan a billion rows in production.
• It does not know your existing indexes, constraints, or partitions unless you tell it.
• It is a pattern machine, not a calculator. Any total, count, or average it states directly can be confidently wrong.
• It happily invents column names, functions, and vendor features that do not exist.

The mental model that keeps you out of trouble: treat AI as a fluent junior engineer who has read every manual but has never seen your database. It writes a fine first draft. It has no idea whether that draft is correct against your data, and it will not tell you it is unsure unless you ask. Every claim it makes about your numbers, your performance, or your schema is a hypothesis you verify, not a result you trust.