Interviewing senior engineers when AI is in the room

Juan Molina ยท

Hiring teams are split on AI in technical interviews. Two 2026 reports show what signals still separate strong senior engineers.

A senior engineer joins a live coding interview, solves the problem cleanly and explains very little. Was that skill, or an assistant on a second screen? Hiring teams now face that question in almost every remote technical loop, and they do not agree on the answer.

Teams are split on the rules

CoderPad's State of Tech Hiring 2026, based on more than 650 developers and recruiters, found no common policy. 34% of respondents said their company bans AI in interviews, 29% allow it with constraints, 17% allow it broadly and 20% decide case by case (CoderPad, February 12, 2026).

Whatever the policy, candidates are using it. Vettasy, a vendor that sells interview monitoring, analyzed 14,260 interview sessions and detected AI assistance in 42.6% of them, rising to 55% in backend roles and 61% in machine learning roles (Vettasy, September 29, 2026). The rate it measured rose from 24.1% to 52.4% across the period studied, with a reported false positive rate of 3.0% (Vettasy, September 29, 2026). Read those figures with care: the source sells detection, and its own detection improved over the period, which can push measured rates up.

What still signals senior skill

The more useful part of the CoderPad data is what respondents said shows real ability when AI is available. 66% pointed to a candidate who catches and fixes mistakes in AI output, and 56% to one who explains the trade-offs behind a solution (CoderPad, February 12, 2026).

That matches what a senior role is. You do not hire a staff engineer to type code faster. You hire them to judge whether code is correct, safe and maintainable, and to explain that judgment to a team. An interview that only checks whether a working answer appears tests the wrong thing.

CoderPad also found a gap between habit and value. 69% of recruiters use resume review, but only 16% say it predicts job performance, and algorithmic puzzle questions are still used by 43% (CoderPad, February 12, 2026). The formats rated most effective were technical discussion, chosen by 56% of developers and 60% of recruiters, and live coding, chosen by 43% of developers and 60% of recruiters (CoderPad, February 12, 2026).

The real risk is not the assistant

Among cheating concerns, 52% of CoderPad respondents named agent-generated solutions and 20% named identity fraud (CoderPad, February 12, 2026). The second one is the more serious problem for a senior hire. A candidate who uses AI well is a candidate who may work well. A person who is not the candidate at all is a security risk.

So the decision is less "ban or allow" and more "what do I actually want to observe?" If you allow AI, design the exercise so the assistant is part of the test: give a flawed generated solution and ask the candidate to review it, or ask them to defend a design choice under follow-up questions. If you ban it, say so clearly and make the session conversational enough that a hidden assistant does not help much.

What to do this week

  • Write down your current AI policy for interviews and share it with candidates before the first technical round.
  • Replace one algorithmic puzzle in your senior loop with a review of flawed, AI-generated code.
  • Add a trade-off discussion where the candidate must defend a design under follow-up questions.
  • Check how you confirm identity at the start of each remote technical round.
  • Ask interviewers to score judgment and explanation separately from whether the code runs.