It’s been roughly three months since “Vibe Coding” entered our vocabulary, and in that brief time, it has undergone a transformation that has stripped it of its original meaning. It has now become just another buzzword, much like “Agile” or “REST APIs” before it.
The Original Definition: A Specific Approach
When Andrej Karpathy, a co-founder of OpenAI and former AI leader at Tesla, first coined the term “Vibe Coding” in February 2025, it had a very specific meaning. Karpathy described it as a new approach where you “fully give in to the vibes, embrace exponentials, and forget that the code even exists.” It meant using AI tools to generate code by simply describing what you wanted, without ever looking at the code itself — just seeing the results and iterating through natural language prompts.
The concept is straightforward: you describe what you want in plain English rather than in a programming language, and the AI (hopefully) generates the code for you. You won’t review the code line by line; you just run it to see if it works. If it does, great. If not, you refine your prompt and try again. This approach was explicitly positioned as “not too bad for throwaway weekend projects” and not intended for production systems.
The concept wasn’t new - Cursor’s Composer was around for a bit - but with the catchy phrase and the well-known author, the tweet was widely shared.
The Hype Cycle Begins
What happened next has become a familiar pattern in the tech industry. Self-proclaimed thought leaders latched onto the term, taking everything out of proportion. Suddenly, Vibe Coding wasn’t just a prototyping technique - it was revolutionary, it was the future, it was going to replace traditional software engineering entirely.
LinkedIn and Twitter were flooded with grandiose claims. “All software engineers will be out of a job in a year!” “Production-ready systems built entirely through prompts!” “The end of traditional coding as we know it!”
These claims were, of course, wildly overblown. But they appeared smart to the uninformed crowd eager to seem forward-thinking on social media.
The Reality Check
As the hype cycle continued, a counter-movement emerged. Not from the typical AI-nay-sayers, but from those who actually build production-grade systems and have been pragmatically incorporating AI tools into their workflows.
These experienced engineers know from first-hand experience that while AI code generation tools can significantly speed up development, they by no means eliminated the need for skilled software engineering. In fact, they argued, using these tools effectively required even more expertise to critically evaluate the generated code, make necessary modifications, and ensure it met production standards.
The reality check was clear: “If you want to use Vibe Coding for anything serious, you actually have to look at the code and understand it.” What works for a quick prototype is absolutely not the same thing as building a production-grade system that is secure and scales reliably (which is also true when humans write the code).
Lost in Translation: Non-Developers
One crucial aspect of the original Vibe Coding vision was its democratizing potential. The promise was that anyone could simply describe what they wanted in natural language, and the AI would create it for them without requiring any coding knowledge.
We’ve seen this vision embodied (before the term Vibe Coding was coined) in tools like Lovable, v0, Bolt, and similar products that genuinely try to make software creation accessible to non-developers.
A Term Has Lost Its Meaning
“Vibe Coding” has been diluted to the point where it’s synonymous with “AI-assisted coding” of any kind. This pattern is not unique to “Vibe Coding”. We’ve seen similar semantic drift with other technical terms:
“Agile” began as a specific methodology with clear principles (“Individuals and interactions over processes and tools”) but has become a vague buzzword that companies claim to follow while often adhering to none of its actual tenets.
“REST APIs” had a specific technical definition outlined by Roy Fielding, but in practice, the term is now applied to almost any HTTP-based API. REST maturity model - what was that again?
The Battle Is Lost
While a part of me wants to fight for the original definition of Vibe Coding, the battle has been lost. The term has been co-opted and diluted and has become a synonym for “AI-assisted coding”, likely because “Vibe Coding” is much catchier. It’s memorable, it’s snappy, and it has a coolness factor that “AI-assisted coding” lacks.
As the proud owner of vibe-coding.cc (originally purchased for an April Fools joke), it’s only to my benefit, right? :)