Why not ask for something so outrageous in scope that only effective wielding of Claude Code could yield it in an hour or so?

I make it clear that AI is not allowed at the remote coding assessment and I enforce it by asking frequent questions on the code they type. It's nearly impossible to maintain a "natural" back and forth conversation when you're copying code you don't understand fully. Of course they can mirror my questions to an AI, but that normally introduces unnatural lag in their answers. I think I rarely get false negatives, and if I do I don't care - if the candidate is able to use an AI and sustain informative communication at the same time, he's undistinguishable from a "good" engineer for all intent and purposes. I don't know how often I get false positives, but again I don't care - if the candidate is not able to sustain a conversation while coding, he's not a remote coworker I want in my team, regardless of whether he's using AI or not.