[lev99]

> A screenshare would be a better representation of on-the-spot problem solving, if that’s what you’re testing, because I’m at least in an environment I typically work in.

I do coding interviews in the following manner: I select an interesting problem from a book. I personally complete the problem, measuring what was challenging and the time it took for me to complete the problem. I check the textbook solution, making notes of how my solution differs from the textbook's. If I like the problem, and if it fits with the development role, I invite the candidate to bring a development laptop, I give them a hard time limit of twice the time it took me to complete it, and have them share their screen while they attempt to complete it.

Afterwards we test their solution, I ask them some questions about it. I grade based on code taste, the number of hints they need, and their completion time. All candidates for a position are given the same question. Performance in coding interview is about two third's of the overall candidate evaluation. The ability to sit down and write good code in a time constraint manner is a very important part of being a developer.

[ordinaryperson]

This would be a good method if your goal is to hire a carbon copy of yourself.

First, your definition of "interesting problem" probably includes a lot of personal biases. What you find interesting probably touches stuff you've had experience with, but just because you've done work on say, customized implementations of binary search, doesn't mean it's a good baseline for all developers everywhere. Maybe they've worked on different domains or solved different problems and can't relate to questions you personally find "interesting".

It'd be much fairer to select a boring question that involves standard stuff that everyone has to face at some point in their careers, like string manipulation. And even then, it just involve normal transformations ('look for and remove special characters') and not wacky, HackerRank-style rules ('no two bs after cs but before as').

[munchbunny]

I mostly agree with this approach. I think that you absolutely should be measuring the "boring stuff" because that's 90% of the work by number of hours spent.

On the other hand, I think there needs to be some proxy for "how well does the interviewee navigate complexity?" Because a programmer who does it well is far more valuable than a programmer who doesn't, and there's high variance on this between programmers and between fields of expertise.

The systems design interview sort of gets at it, and whiteboard interviewing often tries to get at it. I feel like I've yet to settle on an approach I'm happy with.