There are still lots and lots of organizations treating LLM usage during coding interviews as cheating.

I think that needs to change.

If you expect your engineers to work with AI, your hiring process should evaluate how effectively candidates work with AI. That is becoming part of the job.

This is still far from standard practice. In CoderPad’s 2026 survey, 34% of surveyed hiring leaders said AI was not allowed in interviews and assessments. That is a vendor survey, so I would not treat it as a precise measurement of the entire industry. But it gives us an idea of the disagreement.

The candidate still has to understand the code.

I have written before about the problem with blindly trusting AI-generated code. If you are a professional software developer, you are responsible for what you ship. That responsibility continues when an AI tool writes part of it.

The same principle should apply in an interview.

A candidate can generate a feature quickly and still misunderstand the requirements. They can produce code that passes a few tests and still miss a serious failure case. Finishing the task tells you something. You also need to understand how they got there.

So give them a small codebase. Ask them to add a feature. Let them use AI, and make that permission clear before the interview.

Then watch what happens.

Do they clarify the requirements? Do they give the model enough context? Do they read the changes? When a test fails, can they investigate it? Can they explain why they accepted one suggestion and rejected another?

Change a requirement. Ask them to adapt the solution and explain the consequences.

I think those are useful things to evaluate in a software developer.

Some companies have already moved in this direction. Canva publishes criteria for AI collaboration and technical ownership. DoorDash describes a 60-minute AI-assisted engineering session that evaluates verification, debugging, and judgment. In May, Business Insider reported that Google planned a pilot letting candidates for selected junior and mid-level US roles use Gemini during a code-comprehension round, with interviewers assessing prompting, output verification, and debugging.

Fundamentals still matter. You can ask candidates to explain a design decision without consulting an LLM. You can check whether they understand the code they just produced. An interview can evaluate both independent understanding and the ability to use tools effectively.

I do not know whether these formats will predict long-term job performance better. We should measure that too.

But if working with AI is part of the role, I think it belongs in the evaluation.

If you are hiring software developers, give candidates an opportunity to demonstrate that skill. Then hold them responsible for the result, just as you would after hiring them.