In February 2025, Andrej Karpathy — one of AI’s most followed figures — posted a casual description of a coding style he had been using: just tell the AI what you want, accept the output, barely read it, fix errors when they surface. He called it “vibe coding.” Within weeks, the phrase had escaped its original context entirely.
By mid-2025, “vibe coding” was appearing in job ads, startup descriptions, accelerator pitches, and developer conference panels. A term that started as a self-aware joke about letting AI do the thinking had become a genuine professional category — complete with debate about whether it was good practice, a useful shortcut, or a way to produce code nobody understands.
The speed of the transition reveals something interesting about how internet culture now creates legitimate vocabulary. In previous eras, a word needed years — or at least a dictionary editor — to move from slang to professional use. Vibe coding took months. The mechanism was not slow cultural absorption but rapid institutional adoption: companies using the phrase in hiring, media covering it, conferences featuring it, until the irony drained out and only the label remained.
What makes this pattern culturally significant is that the underlying behaviour was not new. Developers have always copied code they didn’t fully understand. The difference is that AI made that approach so much more powerful — and so much more visible — that it needed a name. Karpathy’s phrase arrived at exactly the moment the practice became widespread enough to discuss.
There is a recurring internet dynamic here: a phrase that describes something slightly embarrassing or informal gets named, and the naming makes it legitimate. Once “vibe coding” existed as a term, it was possible to defend it, refine it, argue for its appropriate uses, write articles about it. Before the name, it was just something people did quietly.
The debate around vibe coding — is it real coding? does it matter if you understand the output? what are the long-term maintenance implications for a codebase? — is a genuine and useful one. But that debate only became possible because the culture produced a sticky enough phrase to carry it.
This is what internet culture now does with technical practices that AI makes newly visible: it names them fast, debates them immediately, and mainstreams them before any formal consensus forms. Vibe coding is a particularly clean example, but the pattern will keep repeating as AI tools change more workflows that previously had no names because they were too informal to discuss.
The phrase started as a wink. It ended up doing serious definitional work. That journey, compressed into just a few months, says at least as much about how the internet processes new ideas as it does about AI development.
