The Truth About AI-Assisted Interviews and Why Most Tech Candidates Still Can't Code
By someone who's interviewed hundreds of engineers can see AI-assist from a mile away
There's a lot of talk about Roy Lee and AI-assisted tech recruiting going around. That story spread fast because it touched a nerve. Not just the scale of what he pulled off, but how familiar it all felt. In my career, I've interviewed hundreds of tech candidates, and I've seen first hand the impact that AI-assisted job interviews have had.
This article nailed the core issue: it isn't a candidate problem. It's an interviewer problem. Roy Lee got through because people didn't dig. They accepted the performance. They didn't stay in the conversation.
A Real Conversation
The best interviews don't feel like tests. They feel like working sessions. Two people walking through an idea, figuring out what someone really understands.
When you ask someone to explain how they built something, you're not just checking knowledge. You're checking depth. A surface-level answer might sound fine until you follow up. Ask for more detail. Ask how they handled the edge cases. Ask why they made that call. The more you dig, the more the truth shows up.
Everyone hits the bottom eventually. That's not a problem. That's the whole point. A good interviewer knows how to get there without turning the interview into a hostile exercise.
Where AI Fits
AI won't fix hiring, and it certainly won't replace good interviewers. What it can do is help all interviewers get better.
Used well, AI can suggest sharper questions. It can flag when a candidate's story contradicts itself. It can keep the conversation moving when a human might get stuck. But it shouldn't make the call. That's still on the person doing the interview.
This is how we see it: AI should make people better at doing hard things. Interviewing is hard. Listening is hard. Adapting is hard. The best tools should support the human in the loop to keep those things going, just like Roy Lee's tools did for the candidate side.
And if a candidate is using AI too? From my perspective, that's fine. They still have to understand what they're saying. They still have to explain it. They still have to stand behind it.
Still, Some Can't Code
Even with good questions, there's one truth that never really changes. Some people just can't code.
In my experience, half the candidates applying for developer jobs fall into this category. Not beginners. Not out of practice. They just can't write code.
You don't need a complicated screen or Leetcode to figure that out. A very short live coding session is usually enough. Simple problems, clear expectations. But they need to be able to talk through their work. That's what engineers do day-to-day, after all.
What Matters
There will always be noise in hiring. Always some way for people to slip through if no one's paying attention. But the solution isn't more structure for structure's sake. It's sharper interviewers. More curiosity. Better tools. More real conversations.
Roy Lee wasn't the problem. He just showed us where the cracks are.