Although we don’t have to do LeetCode interviews anymore, now we have AI coding interviews—and they’re not necessarily any easier. They’re even more unpredictable.

The following is from a Reddit post

-------

To everyone who thinks AI interviews are great: good luck with that :) Bloomberg interview experience [London/Frankfurt]

I just finished the second round of Bloomberg’s AI codebase interview.

It tested Python’s Django framework, which I knew absolutely nothing about. And the codebase was enormous.

They spent the first 30 minutes on the behavioral interview, so you basically only had 30 minutes left to write code.

The boundaries around using AI tools were incredibly vague. You had no idea how much AI the interviewer would allow you to use, and they wouldn’t clearly state their expectations. I used planning mode and ask mode, working through it feature by feature. I passed a few test cases, but still got rejected in the end.

When you’re using AI assistance, you have to read a huge amount of generated content. Trying to keep up a conversation with the interviewer at the same time makes it practically impossible.

When we used to grind LeetCode, at least the problem was clear: you just had to figure out how to solve it. Even if you got stuck, you could brainstorm different approaches. But in these stupid AI interviews, you’ve already spent half your time just trying to understand the huge codebase and figure out what’s wrong.

To those who were celebrating the “end of the LeetCode era”: good luck. Interviews are only going to get much harder now.

Oh, and don’t even think about “one-shotting” the whole problem with a single prompt to AI, or you’ll still get rejected (that happened to my friend).

Anyway, AI interviews are complete bullshit. There’s no way to find a reliable, universal approach to handling them.