Why LeetCode Was Always a Dumb Idea - And AI Is Finally Killing It
LeetCode never tested real engineering skills, and AI has finally exposed its fundamental flaws as a hiring filter
Why LeetCode Was Always a Dumb Idea - And AI Is Finally Killing It
I've never bought into the LeetCode cult. For years, I watched (and yeah, sometimes joined) the madness: devs like me sacrificing nights and weekends to grind hundreds of algorithm problems - two-pointer this, DP that, sliding window everything - just to survive a 45-minute interview. We told ourselves it was worth it. It wasn't.
Look, in the real world I've never once needed to reverse a linked list in my head or optimize some trie while a stranger stares at me. What I do need is to write code that other humans can read, debug a production outage at 3 AM, argue about architecture in a meeting, and ship features that actually matter. LeetCode didn't test any of that. It tested how well you could memorize patterns and perform under artificial stress. Great for competitive programming contests, terrible for spotting who'd actually be a good teammate.
And the side effects sucked. An entire cottage industry sprang up - $500 courses, premium subscriptions, Discord grind groups - turning interview prep into a second job. It burned people out, favored those with endless free time, and quietly filtered out talented engineers who just weren't wired for timed puzzle battles.
Then AI rolled in and laughed in its face.
By late 2024, models like ChatGPT and Grok were casually crushing mediums and even a lot of hards. Tools popped up that could feed you solutions during virtual interviews faster than any human could type. The emperor had no clothes, and suddenly everyone could see it. If an AI can solve your precious puzzle in seconds, what the hell were you actually measuring? How fast someone can pretend it's 1995 and Google doesn't exist?
Companies panicked, of course. Some doubled down with harder problems or dragged everyone back onsite. Others quietly started ditching the LeetCode gauntlet for take-homes, pair programming, or straight-up real-world tasks. The "no LeetCode" company lists got longer, and honestly? Good. It's about damn time.
The end of an era
LeetCode isn't dead yet - some big tech dinosaurs still cling to it like a security blanket - but its days as the golden ticket are over. AI didn't break it; it just exposed how broken it always was.
My take: stop torturing yourself with endless grind sessions. Build stuff you're proud of, ship open-source code, and get really good at using AI as your pair programmer. That's what actually moves the needle now.
LeetCode had a decent run, but it was a flawed, overhyped filter from the start. I'm not sad to see AI drag it to the grave. The sooner it's gone, the better hiring - and coding - will be.