Courses › AI-Assisted Coding › The New Round
Reviewing AI Code
Catching the subtle bugs assistants introduce, and testing as you go.
AI assistants produce code that *looks* right. The bugs they introduce are often subtle — the kind that pass a quick glance and a happy-path test. In an AI-assisted interview, catching them is the clearest signal of engineering skill.
Bugs assistants commonly introduce
A review checklist
Before accepting a change, ask:
- Does it do what I asked — and only that? Look for unrequested changes elsewhere.
- What are the inputs, and what happens at their edges? Mentally run empty, one, many, invalid.
- Do all the called functions really exist with these signatures?
- Does it follow the codebase's conventions for errors, logging, naming and structure?
- What's the complexity for realistic input sizes?
- Do the tests check the spec, including failure cases?
Testing as you go
- Run the existing test suite after every change — catch regressions immediately.
- Add a test for each edge case you identify, *before* asking for a fix.
- For logic you didn't write, a quick manual example often reveals more than reading: "let me run this with an empty list and with duplicate due dates."
Debugging with an assistant
- Reproduce the failure and read the actual error yourself.
- Form a hypothesis — say it out loud.
- Narrow it down with a minimal case or a print statement.
- Then use the assistant to help explore or implement the fix — with your hypothesis as context.
[!WARNING] Never "fix" a test to make it pass If a test fails, the code is probably wrong. Changing the assertion to match buggy output — something assistants will happily do if asked to "make the tests pass" — is a serious red flag to an interviewer.
[!INTERVIEW] Call out what you catch When you spot a problem in generated code, say so: "It's comparing due dates as strings, which only works because they're ISO formatted — I'll parse them to dates to be safe." Catching and explaining mistakes is exactly the signal these rounds exist to find.