Vibe Coding is Dead, Long Live Vibe Coding
What my all-caps AI meltdown taught me about the new rules of coding.
Early 2010s: Everyone had an app idea. The App Store was blowing up and suddenly every third person was a wannabe tech entrepreneur. I remember friends, colleagues, dinner party guests, dropping their elevator pitches, sometimes even jotting down notes on cocktail napkins, perfectly capturing the Zeitgeist — the vibe — of the moment.
Back then the bottleneck wasn’t vision, it was technical talent. You needed someone who could build it and that person was suddenly the most important person in the room. There's good reason why so many of today’s billionaires were once coders.
Today, exponential advances in AI have delivered the nirvana of vibe coding: Describe what you want, let the model write the code, and bam, you’ve got your dream product.
Have a big idea? No need to pitch your techie friends. Lovable and Cursor are serious platforms with great UIs, making not just coding, but delivery, easy. Need a website? Wix and Framer AI are at the forefront, making a decent looking website with a few prompts and some templates thrown in.
2013 me: eat your heart out.
Alas, like the 2010s, there are problems: Unmaintainable codebases. Security holes nobody caught. Mystery features nobody asked for. Codebases that worked great until they really, really didn't.
Rather than “prompt-and-pray”, today vibe coding requires professional discipline: Architecture decisions made by humans. Functional requirements. Design ideas. Code review and QA wrapped around every PR.
My first experiment, a browser screenshot diff tool, crashed and burned after many glaze-eyed late-night sessions, culminating in an expletive-filled melt-down: the library "we" built everything around wouldn't work in a web app. OMG YOU @#$%@ WASTED A WEEK OF MY LIFE! The human was asleep at the job.
Vibe coding hasn’t so much died as evolved. Talent is still needed: Good ideas with market fit. Great sense of design and UX. Focus on the features that matter. And, of course, building stuff that not only works, but that works with edge-cases, high server load and bad actors trying to ruin your day.
Boris Cherny, creator of Claude Code, doesn't even call it vibe coding anymore. It's just coding. On a recent Lenny's Podcast episode he describes how designers, PMs, even finance people are building at Anthropic. Engineers? They’re conducting chaotic symphonies of agents.
His framing is telling: old-school punch-card programmers once insisted software languages weren't real coding. Every generation draws the line somewhere.
The difference? Rather than taking the original “prompt-and-pray” approach, today vibe coding requires professional discipline: Architecture decisions made by humans. Functional requirements. Design ideas. Code review and QA wrapped around every PR.
The need to ship fast without breaking things is why we built ReleasePass, our web QA platform. It leverages both heuristics and AI to find defects that would take a human hours across multiple tools. My first vibe-code misadventure shaped the process: ReleasePass was obsessively researched and documented. No oh-s*** moments.
For many companies we work with, “we vibe coded it" lands either with derision or street cred, depending on where they are on the narrative arc. The question worth asking your dev partner: what's your process around AI output? Make sure they have a good answer.
The freewheeling cowboy era is over. AI still needs human judgment for architecture, for design, for knowing what good looks like. The value didn't disappear. It just moved up the stack — maybe where it always belonged.