The even bigger challenge is that hiring experts in any domain requires domain knowledge, but hiring has been shifted to HR. They aren't experts in anything, and for some years they made do with formulaic approaches, but that doesn't cut it anymore. So now if your group wants to get it done, and done well, you have to get involved yourself, and it's a lot of work on top of your regular tasks. Maybe more work because HR is deeply involved.

> Company A wants to hire an engineer, an AI could solve all their tech interview questions, so why not hire that AI instead?

Interview coding questions aren't like the day-to-day job, because of the nature of an interview.

In an hour-long interview, I have to be able to state the problem in a way the candidate can understand, within 10 minutes or so. We don't have time for a lecture on the intricacies of voucher calculation and global sales tax law.

It also has to be a problem that's solvable within about 40 minutes.

The problem needs to test the candidate meets the company's hiring bar - while also having enough nuance that there's an opportunity for absolutely great candidates to impress me.

And the problem has to be possible to state unambiguously. Can't have a candidate solving the problem, but failing the interview because there was a secret requirement and they failed to read my mind.

And of course, if we're doing it in person on a whiteboard (do people do that these days?) it has to be solvable without any reference to documentation.

One of the best interviews I've encountered as a candidate wasn't exactly a pair programming session but it was similar. The interviewer pulled up a webpage of theirs and showed me a problem with it, and then asked how I would approach fixing it. We worked our way through many parts of their stack and while it was me driving most of the way we ended up having a number of interesting conversations that cropped up organically at various points. It was scheduled for an hour and the time actually flew by.

I felt like I got a good sense of what he would be like to work with and he got to see how I approached various problems. It avoided the live coding problems of needing to remember a bunch of syntax trivia on the spot and having to focus on a quick small solution, rather than a large scalable one that you need more often for actual work problems.

Problem is, company A doesn't need an engineer to solve those interview questions but real problems.

Tech interviews in general need to be overhauled, and if they were it’d be less likely that AI would be as helpful in the process to begin with (at least for LLMs in their current state).

Current LLMs can do some basic coding and stitch it together to form cool programs, but it struggles at good design work that scales. Design-focused interviews paired with soft-skill-focus is a better measure of how a dev will be in the workplace in general. Yet, most interviews are just “if you can solve this esoteric problem we don’t use at all at work, you are hired”. I’d take a bad solution with a good design over a good solution with a bad design any day, because the former is always easier to refactor and iterate on.

AI is not really good at that yet; it’s trained on a lot of public data that skews towards worse designs. It’s also not all that great at behaving like a human during code reviews; it agrees too much, is overly verbose, it hallucinates, etc.

I want to hire people who can be given some problem and will go off and work on it and come to me with questions when specs are unclear or there's some weird thing that cropped up. AI is 100% not that. You have to watch it like a 15 year old driver.

A company wants to hire someone to perform tasks X, Y and Z. It's difficult to cleanly evaluate someone's ability to do these tasks in a short amount of time, so they do their best to construct a task A which is easy to test, and such that most people who can do A can also do X, Y and Z.

Now someone comes along and builds a machine that can do A. It turns out that while for humans, A was a good indicator of X, Y and Z, for the machine it is not. A is easy for the machine, but X, Y and Z are still difficult.

This isn't a sign that the company was wrong to ask A, nor is it a sign that they could just hire the machine.

It's because coding interview questions aren't so much assessing job skills as much as they are thinly veiled IQ tests.

I think if it was socially acceptable they'd just do the latter.

This is a great point. Though what if the answer is that the company can hire that AI to solve a significant fraction of its actual problems? People who do the assessments and decide what features should look like are often called managers (product, engineering, etc.).

For a while I’ve been skeptical that the rate of hiring of engineers would change significantly because of LLMs, but I’m starting to feel like maybe I’m wrong and it’s already changing and companies are looking toward AI to lower costs and require fewer humans. In that case they are probably still going to want people who are technically exceptional - maybe even more so - but are able and willing to create, integrate, and babysit AI generated code, and also do PM and EM style feature management.

If companies are slowing hiring due to AI, I would expect interviews to get worse before they get better.

> For example, it could be that the company wants the engineer to do some kind of assessment whether a feature should be implemented at all, and if yes, in what way. Then you could, in an interview, give a bit of context and then ask the candidate to think out loud about an example feature request.

So a Product Manager?

[dead]