Interviews in the Age of AI
by Gordon Weakliem
I think one of the early outcomes of the AI revolution is that Leetcode interviewing is dead. Sure, there are countermeasures, you can make them share a screen or swear that they won’t use AI, but is that how you want them on the job? The point of the interview is to get a sense of how they are as a co-worker.
About 20 years ago, I used to do interviews that were “write a program to add two numbers together” (I think specifically I asked for a web application). It’s trivial, right? There’s actually a lot going on. You have to parse the input and maybe there’s garbage in the input, or empty input. Fixed-size integers can overflow. Your marketing team can suddenly decide that you need a four-function calculator, not just addition. Ask the candidate to write some tests. Ask about how they’d handle arbitrarily large numbers.
What I liked about that process is that it relied less on their ability to suss out a solution to some problem they’ll never have to solve on the job and focused more on average activities. Sometimes a candidate will ask a lot of questions is an opportunity to see them refine requirements.
Another way to do this is to give the canditate a PR with some issues and ask them to give feedback on the PR. This is a way to see how they give feedback - is it just nitpicky stuff or are they identifying real issues? It can also reveal how they think about code - do they look at something with a bunch of repetitive, copy/paste stuff and think it’s fine like that?