Killing the Leetcode-style coding interview is either a brave step into reality… or a convenient way to stop measuring the one thing that’s easiest to measure.
Anthropic reportedly deleted the classic algorithm puzzle round from its interview loop because AI has made it pointless. And honestly, the headline is right in one blunt way: if a candidate can quietly lean on an AI assistant to crank out a clean solution, then the interview stops being a test of skill and starts being a test of vibe, setup, and how good someone is at hiding help. That’s not hiring. That’s theater.
But I don’t think the real story is “AI beat Leetcode.” The real story is that companies are finally admitting something they benefited from for years: those interviews were never a great proxy for the work most engineers do.
Leetcode rounds had one big advantage. They were standardized. You could compare candidates fast. You could claim it was “merit-based.” And you could run it at scale without needing a deep, time-heavy evaluation. The problem is what they actually rewarded: recent practice, comfort under pressure, and familiarity with a specific genre of problem. Those things are not nothing. But they’re also not the job for a lot of roles.
AI just made the mismatch impossible to ignore.
If you’re hiring someone today, you’re not hiring them for their ability to recall a trick for reversing a linked list in 20 minutes on a shared screen. You’re hiring them to ship code in a messy codebase, read other people’s decisions, make tradeoffs, and not break production. And increasingly, you’re hiring them to work with AI tools as part of their normal workflow. So testing a person in an artificial environment where they can’t use the tools they’ll use on day one is starting to look… silly.
Still, removing Leetcode doesn’t automatically mean hiring gets better. It can get worse.
Because the moment you remove a standardized filter, you replace it with human judgment. And human judgment is where bias lives. Not always malicious bias—often the casual kind. “I liked them.” “They seem sharp.” “They remind me of me.” That’s the risk. If Anthropic (or anyone) replaces puzzles with looser conversations, they might accidentally make hiring more about storytelling and confidence than actual ability to deliver.
And yes, that can be a feature, not a bug, depending on who’s doing the hiring.
Imagine two candidates. One is quiet, methodical, and a little awkward on calls. The other is smooth and fast, with strong opinions and great verbal energy. In a world without concrete tests, the second person can win even if the first person writes better code and makes fewer mistakes. That’s not a rare outcome. That’s the default unless you design against it.
On the other hand, the Leetcode world was also unfair—just in a different direction. It punished people with jobs, kids, or limited time to grind. It favored the people who learned the “meta.” It rewarded memorizing patterns. You could be a great engineer and fail because you blanked on a trick you’d never use again.
So I get why a company would pull the plug. And I even think more should.
But here’s where it gets uncomfortable: AI doesn’t just “make Leetcode pointless.” AI also makes a bunch of other screening methods shaky.
Take-home projects? AI can write a shockingly decent first draft. System design interviews? AI can generate architectures and tradeoffs that sound convincing. Even code reviews can become a performance: someone asks an AI to produce comments that look thoughtful. The new problem isn’t “people can cheat.” People have always been able to cheat. The new problem is that “cheating” now looks a lot like “working normally.”
So what are you actually trying to measure?
If the job is “use every tool available to ship good work,” then using AI in an interview should be allowed. Maybe even required. But then you have to judge something harder: how well they steer the tool, how they verify outputs, whether they catch subtle bugs, whether they can explain why a solution is safe, and whether they know when not to trust it.
That’s real skill. And it’s not captured by a puzzle.
Here’s a concrete scenario: you hire someone who demos well with AI in the interview. They prompt cleanly, produce good-looking code, and explain it confidently. Then they start the job and quietly ship brittle changes because they don’t truly understand the system. They move fast, break things, and leave a trail of “AI wrote it” code that nobody wants to own. The team pays the price months later.
Another scenario: you reject someone because they’re not flashy in conversation, but they’re exactly the kind of person who would have written careful, boring, reliable code for years. The kind of engineer every team says they want, right up until they’re sitting in the interview room.
So I like the instinct behind Anthropic’s move. I think it’s an honest admission that the old test was drifting away from reality. But the hard part isn’t deleting Leetcode. The hard part is building something better that doesn’t just turn into a personality contest.
Because if we get this wrong, we don’t just change interview questions. We reshape who gets hired, who gets locked out, and what kind of engineering culture wins.
If AI is now a normal part of coding, what should an interview actually prove about a person?