Mastery · Track: data engineering · lesson 9.1 open

The risk of rust

How AI dependency rusts data reasoning without you noticing, and a five-task test to measure it.

Promise: a short test that shows whether your skills still work without an assistant.

The phenomenon

Coding assistants speed up work. They can also replace the piece of reasoning you need to exercise: reading an execution plan, writing a SQL window function, modeling a table. If the tool always handles that part, it rusts, and you only notice when it is too late: in an interview, during an incident without tool access, or when someone asks “why does this work?”.

This is not an argument against AI. It is an argument for knowing when you use it and what you still do on your own.

Warning signs

• You accept the solution without running the execution plan.
• You cannot explain a snippet you “wrote” yesterday.
• Technical interviews feel harder than they should.
• On a call, you ask to “look up” things that were basic two years ago.
• You validate the output just by checking if it “looks right”.

Five-task test (45 minutes, no AI and no search)

1. Write a window query that returns the latest record per key (lesson 5.2, pattern 1).
2. Write a sessionization query (pattern 3).
3. Explain out loud why a groupBy can have skew and provide two fixes.
4. Design the model for a fact table in a domain you know, declaring the grain.
5. Describe a CDC → Silver pipeline using MERGE and how it handles deletes.

Score from 0 to 2 (0 = could not do it, 1 = completed with errors, 2 = completed cleanly). Total:

Checklist

• Did I take the test without help?
• Do I have the score and date recorded?

Lessons cited by number that are not in the library yet open later.

Checked on 05/10/2026. Educational content; tax and legal: confirm with a professional.