Back to all articles
Candidate guides

"You Can Use AI in This Interview": What Is Actually Being Scored

When an interview lets you use AI, how you work with it is assessed. The four things the report looks at, the three exercise types, and how to prepare.

H
The Hyrr team
•4 min read

More technical interviews now include a question where you are allowed, or expected, to use an AI assistant. That raises an obvious worry: if everyone has the same AI, what is the interviewer looking at?

In Hyrr's AI exercises the answer is written down. Using the AI is the exercise, not cheating. What gets assessed is how you work with it, under one skill on the scorecard called AI Fluency. The report looks at four things:

  • Delegation: what you handed to the AI and what you kept for yourself.
  • Description: how clearly you told it what you needed.
  • Discernment: whether you caught what it got wrong.
  • Diligence: whether you checked, explained and owned the result.

Counts are recorded (how many problems you found, how many hidden cases your instructions passed), but a count is never the verdict on its own. The report has to explain what the number shows.

There are three kinds of exercise. Each one leans on a different part of that list.

1. Catch the AI: an AI wrote this, so is it right?

You get a piece of work an AI assistant produced, and you review it as if it were about to be used for real. You click a line to flag a problem, then say what is wrong and why it matters.

What it is listening for: discernment. The work has problems planted in it. The interviewer you are talking to does not know where they are and will not tell you "good catch", so silence is not a hint either way.

How to prepare

  • Read it the way you would read a teammate's pull request before it ships, not the way you would mark homework.
  • For each flag, give the consequence ("this drops the last row", "this exposes the key in logs"), not just the symptom.
  • Quality beats count. Five vague flags are worth less than two precise ones.
  • Think out loud while you read. Your reasoning is part of what is heard.

2. Work with AI: use it like you would at work

You solve a task with an AI assistant in the panel beside you. You can use it however you like: to plan, write, explain or check. It can make mistakes.

What it is listening for: delegation and diligence. Code you insert from the assistant stays marked as the assistant's, and everything you send it is part of the interview record. The question is not whether you used it. It is whether the final result is right and whether you can explain it.

How to prepare

  • Decide what you are giving it before you type. "Write the parser" and "explain why this regex fails on empty input" are different jobs.
  • Run or check what it gives you before you build on it.
  • Be ready to say, in your own words, why every line you kept is there.
  • If you threw something out, say why. That is evidence too.

3. Brief the AI: write the instructions, not the answer

You write the instructions an AI will follow to do a task on its own. You can try them on sample inputs, see exactly what the AI answers, and improve them between a limited number of test runs. When you submit, your final instructions are scored on cases you have not seen.

What it is listening for: description. Instructions that only work on the samples will fail on the hidden cases.

How to prepare

  • State the goal, the allowed outputs and the edge cases. Spell out what to do when the input is ambiguous.
  • Use your test runs to find where it breaks, not to confirm it works.
  • Do not tune for the samples. Ask yourself: "What input would trip this up?"

Some employers also set a fourth type, Build an agent: you write an agent's instructions and choose which practice tools it may use. It is scored the same way, on hidden cases.

Two things that hold everywhere

Know which questions are AI questions. An AI exercise opens with a screen that tells you what is allowed. Everywhere else, treat the interview as AI-off unless you are told otherwise. If you are not sure what an interview will include, ask the recruiter before the day.

Talk. In every format above, what you say about your choices counts alongside what you produce. A correct answer you cannot explain is weak evidence. A partly wrong one where you named the risk is better evidence.

Practise before the real one

The Hyrr candidate portal has a practice page with all three exercises: Catch the AI, Work with AI and Brief the AI. You see where you stand straight away and can retake any of them. Pass all three and you can claim a verified AI Fluency badge. It sits with your verified skills, where an employer can check it for themselves.

Practise working with AI


Do you interview engineers at work? Hyrr runs the first-round interview for your open roles, out loud, and gives your team a report on every candidate. Create an employer account

See a live AI interview

Pick a sample role and take the interview yourself, in the browser or by phone. Your scored report lands in your inbox.