Stopping AI in live coding interviews

Of all interview formats, live coding gives AI the easiest job. The problem sits on the screen in writing, and current models produce good code. What follows is how to stop AI in live coding interviews, or at least to slow it and bring it into view.

The ways candidates get AI help with code

• An interview assistant captures the problem off the screen and presents a solution, in some cases inside a window invisible to screen sharing. Interview Coder was one such tool and was the forerunner of Cluely.
• A phone has a chatbot open.
• The code editor has AI suggestions built in, or a browser extension supplies them.
• The problem comes from a public practice site, where solutions are published.

Measures that work in a coding round

Pick problems the tools handle poorly

• Set problems you wrote yourself. Real work from your team is the best source.
• Hand over existing code, so reading it is the first task.
• When the first solution is done, add a constraint: a memory limit, say, or a bug report.

Have the candidate talk through the work

Before any code is written, ask for the plan. During it, ask the reason for each step. Afterward, ask what an empty input would do. Choices that were copied are choices the candidate often cannot justify.

Pay attention to how code arrives

The warning signs are correct code landing in big blocks after little typing, variable names the candidate cannot account for, a lengthy silence and then a finished solution, and a gaze that keeps going to a different area of the screen. Plenty of people think first and type second, so use these as prompts for questions about the code. More in signs a candidate is using AI.

Disable AI in the shared editor

If your shared coding tool has AI completion, switch it off, and have the candidate type the code in that tool. That closes one route. A second device stays open.

Frequently asked questions

Is a total ban on AI in coding interviews the right policy?

Not for every team. Some permit AI for one round on the grounds that their engineers work with it daily. Teams that do so should say it beforehand and reserve one round or more for unaided work.

Are take-home coding tasks still useful?

On their own, less so, because AI finishes them with ease. Pair each one with a live session where the candidate walks through the code and modifies it.

Is it time to return to in-person coding interviews?

In-person rounds do the job at a cost in speed and money. The trade-off is discussed in should you go back to in-person interviews? The wider measures are in how to stop AI use in interviews.