Since Andrej Karpathy’s tweet blew up in February 2025 —4 million views and counting— vibe coding has been the tech world’s shiny new toy.

There's a new kind of coding I call "vibe coding", where you fully give in to the vibes, embrace exponentials, and forget that the code even exists. It's possible because the LLMs (e.g. Cursor Composer w Sonnet) are getting too good. Also I just talk to Composer with SuperWhisper…

— Andrej Karpathy (@karpathy) February 2, 2025

Tools like Cursor and SuperWhisper let you chat with AI, say “build me a game”, and bam, you’ve got a prototype by Sunday night. It’s a fun weekend hack, but it’s nowhere near ready for the real world. Let’s break it down.

What’s This “Vibe Coding” Hype?

It’s simple: you tell an AI your idea—“make a retro app” or “fix this messy UI” —and it churns out code while you chill. Karpathy’s line, “see stuff, say stuff, run stuff, copy-paste stuff,” sums it up. You feel like a coding wizard, slapping together apps in hours. I get the appeal—I’ve messed with it myself. But it’s cool until you realize that AI doesn’t care if it actually works tomorrow. That’s where the cracks show.

Why I’m Calling BS on the Revolution

I’ve led teams through enough production fires to know what holds up—and `vibe coding" doesn’t. Here’s why it’s more hype than help:

• It’s a Mess Under the Hood: The code looks fine until you dig in. I’ve seen AI spit out stuff that’s insecure—take Lovable’s generated code, where sloppy handling of Supabase API keys left tokens exposed for anyone to snatch, ripe for abuse. Good luck finding the bug when you didn’t write it.
• Guessing Isn’t Engineering: You tweak until it “works”, but that’s not a system—it’s a gamble. I’ve had to clean up worse disasters from less.
• Scalability? Nope: That quick app is a nightmare to maintain. I once inherited a codebase so tangled it took weeks to unravel—AI-generated code feels like that on steroids.
• You Get Lazy: Lean on AI too much, and you’re screwed when it flakes. The AI won’t save you when the client calls at 2am.

Prototypes Are Not Products

The LinkedIn crowd loves saying “vibe coding” makes everyone a 10x engineer. Sure, for a prototype. But a prototype isn’t a product, and a product isn’t a business. This creates a shortcut culture that there is no benefit at all unless for the ones hyped with it. I’ve watched startups crash because they thought a weekend hack could scale. Real engineering—systems that don’t buckle, teams that ship—takes depth, not vibes.

My Advice: Play, Don’t Pray

If you’re into vibe coding, have fun—but keep it in the sandbox:

• Hack away on side projects or proofs-of-concept.
• Treat the output like a sketch—review it like your job’s on the line.
• Don’t even think about production without a real engineer’s eyes on it.

To be clear: I’m not anti-AI; I use it to brainstorm or speed up grunt work. But I’d rather trust my team and hard-earned skills than bet on a tool that’s all flash, no substance.

Conclusion

If you’ve made it this far, hats off—you’ve waded through my take on vibe coding.

It’s a slick gimmick for a quick prototype, but don’t be fooled: it’s nowhere near ready for the real stuff—solid systems and businesses need more than AI vibes.

I’ve laid out why I think it’s overhyped and risky, so if you’ve got a “vibe coding” win or wipeout, I’d love to hear it.

Drop your thoughts on LinkedIn or Twitter/X or subscribe to Tech Trenches for more from the trenches.

Happy coding!

{C3P}