Interviewers now often ask how you use AI coding tools. They want to know whether you use them well, not whether you use them at all. A strong answer has three parts.
The three parts#
| Part | What to say |
|---|---|
| Which tools, for what | Name the tool you use for each kind of task, and tie it to a project you actually built. Say what you used AI for and what you kept manual. |
| Where you do not trust it | Security-critical paths, authorization logic, complex async code, and anything without tests to catch mistakes. |
| One time you caught it being wrong | What it generated, what was wrong, and how you caught it. |
The third part matters most#
A specific example proves that you actually review the output. Prepare it before the interview. A good example names the failure (for instance an invented API, a missing edge case or a test the agent edited) and the check that caught it: a test, a diff review, or running the code.
Example answer shape#
- "On my last project I used an IDE agent for boilerplate and tests, and a chat assistant to learn an unfamiliar library. I wrote the authorization logic myself."
- "I do not trust it on security paths or complex async code, or anywhere without tests."
- "Once it generated a pagination query with an off-by-one in the offset. A test for the last page caught it, and I fixed the bound."
Fill in your own project and your own example. A made-up example falls apart under one follow-up question.
Short line#
Say which tools you use for which tasks on a real project, where you do not trust them (security, authorization, complex async, untested code), and one specific time you caught the tool being wrong and how.
Key takeaways#
- Tie tools to tasks on a project you actually built, and say what you kept manual.
- Name where you do not trust them: security, authorization, complex async, untested code.
- Have one concrete example of catching a mistake: what, why it was wrong, how you caught it.
- The example is the most important part, because it proves you review the output.