Why are interviews harder than the job?

The alternative I prefer isn't any of the given ones, but a scaled interview. I start with something that you can solve with a Python one-liner, if you know what you're doing, but if you solve that instantly I've got a series of questions that will scale up until we're eventually doing the task using untrusted user input, outputting into a security-sensitive environment, under heavy load that we want to be cheap, in a highly-available environment, etc. etc. I also tell the applicant up front that this is the plan, because if you don't do that, it feels like you're jerking them around by constantly changing the requirements. I've also got a number of directions it can go depending on the applicant, and depending on what they tell me about their past experience.

The other nice thing about this is your interviewee generally doesn't leave feeling like a failure; it's not like I have three questions and you can get them all wrong. Unfortunately there are some people who end up spinning on the very easiest question for the entire session, and, uh, well... I can only do so much, if you really don't know anything about programming at all. This is at least the exception, though.

I have not had to do an interview in the age of practical AI yet, though. In person I don't think I'd have to change much, I've always interviewed with a policy of "I'm not worried about whether the string split command takes its parameters in this or that order, I just want to know you know it exists" and I can basically serve as an AI in the same way I was already being the API reference. Remotely, I'm not sure what I'd do yet.