How do you use AI coding tools in your daily work, and where do you not trust them?

How to answer

Some interviewers will have watched you use the tools. In an April 2026 post on its engineering interview, Sierra says candidates build with AI tooling of their choice in its onsite, and that the review digs into how they used AI along the way. ^[1]Source 1The AI-native interviewPublisherSierraSource typecompany blog OpenAI’s interview guide says some formats allow AI tools and others assess problem-solving without them. ^{[2, 3]}Source 2Interview guidePublisherOpenAISource typecompany hiring pageSource 3OpenAI interview guide (Wayback Machine capture)PublisherOpenAI (archived by Internet Archive)Source typearchived company page Your daily-work answer is the same either way; the lesson on AI rules by company and stage covers the round rules.

The interviewer wants judgment, so give them a map with edges.

1. Where it earns its place, by task. Two or three concrete uses: reading an unfamiliar codebase, drafting tests, scaffolding an integration against a documented API. Say how you drive it, such as writing the interface and tests first and letting the agent fill in the rest.
2. How you check its output. You run it, read the whole diff, check each external call against the documentation for the version in use, and, where the customer’s terms allow, try it on their real system before trusting it. Review a teammate’s agent-written diff the same way, starting from the tests: what do they actually assert?
3. Where you switch it off, and why. Authentication, permissions and secrets; migrations against production data; code you could not judge if it were wrong; anything the customer’s data terms forbid sending to a third-party model. Know what you do when the customer bans third-party models: a model they host, or no tool at all.
4. One time it was wrong. The plausible output, the check that caught it, and the habit you changed.

The trap is the vague middle: “it’s a productivity boost, but I always review the output.” That claims judgment without showing any. The interviewer believes the line you draw only when you say where it is and how you found it.