A twist happened when they announced that using LLMs in a pair programming coding interview round was permitted. I welcomed that news and thought to myself, "Well, that's a first. They must be true to their AI pivot." I skipped LeetCode prep, as LLMs are faster at implementing LeetCode questions than I am at my top typing speed.

I analyzed what else it's possible to spend an hour pair programming on if an LLM is solving the menial coding. I've had some pretty good ideas about what excellent signal can be extracted from a coding interview without the coding part.

Uplifted, I went into an interview and was given a straight LeetCode question that I hadn't practiced. I attempted to use an LLM but was stopped, "because it would just give you an answer." I was permitted to ask it anything, but nothing that related to the coding problem itself.

I spent a minute trying to understand what use I could get out of an LLM with these constraints and decided to wing the problem myself instead. I failed the interview.

In my opinion, the only credible outcome from asking candidates that question is if they have seen and done it before. It becomes significantly easier if you have.

The choice of the coding question also suggests they are not differentiating coding rounds between different pipelines.

The pair programming part that was heavily advertised was absent as well. It was just a standard coding interview.

I don't mind; companies are free to select candidates in a way they see fit. I do, however, dislike being misled in my preparation efforts, especially since this round was advertised so uniquely. I was working on other skills that are not used during standard coding rounds.

Standard coding question for SWE, less common for SRE.