I don't think this argument is any different to the arguments for using libraries, for googling things or looking at Stack Overflow, or against interviews that go too hard on the data structures and algorithms. All of these move the needle from theoretical to practical, and likely give you a better idea of how the candidate will perform on day 1.
The problem however is not day 1, it's that tricky bug that takes one person a few days to solve, but causes another to give up and rewrite the codebase in the next framework. Or it's a slow decline into performance hell, vs constant attention to performance improvements.
I've seen candidates who were very capable React engineers but who, when presented with a pure Javascript problem outside of a React context, were unable to even start writing code. I suspect with AI they would have managed the fairly trivial problem, but I would have lost that signal, and they would still not be comfortable doing general purpose programming outside of React UI code. In other words, AI wouldn't make them qualified for the job, but it would get them through the interview.
Based on this, I think I'd rather candidates didn't use it, or it was limited to "good tab complete" and fixing typos (which does slow candidates down and provides little signal). If AI gets good enough to be able to write entire codebases then sure, let's use them in interviews more, but also I'd probably skip programming interviews and focus on architecture work.
I don't think this argument is any different to the arguments for using libraries, for googling things or looking at Stack Overflow, or against interviews that go too hard on the data structures and algorithms. All of these move the needle from theoretical to practical, and likely give you a better idea of how the candidate will perform on day 1.
The problem however is not day 1, it's that tricky bug that takes one person a few days to solve, but causes another to give up and rewrite the codebase in the next framework. Or it's a slow decline into performance hell, vs constant attention to performance improvements.
I've seen candidates who were very capable React engineers but who, when presented with a pure Javascript problem outside of a React context, were unable to even start writing code. I suspect with AI they would have managed the fairly trivial problem, but I would have lost that signal, and they would still not be comfortable doing general purpose programming outside of React UI code. In other words, AI wouldn't make them qualified for the job, but it would get them through the interview.
Based on this, I think I'd rather candidates didn't use it, or it was limited to "good tab complete" and fixing typos (which does slow candidates down and provides little signal). If AI gets good enough to be able to write entire codebases then sure, let's use them in interviews more, but also I'd probably skip programming interviews and focus on architecture work.