Coding Vibes: Revolutionary Trend or Just Another Buzzword?

In recent weeks, a new trend has been shaking the foundation of the tech innovation field. It involves dictating natural language instructions to a Large Language Model (LLM) which then generates lines of code to create a functional program. This is done without fully understanding the underlying mechanisms, relying more on intuition than the traditional rules of programming. This method, known as vibe coding—or “coding by instinct” in English—“is becoming popular in the tech community,” noted the specialized media outlet Ars Technica in February. Is it a revolution, or just another fleeting buzzword?

The Art of “Saying Stuff, Launching Stuff, and Copy-Pasting Stuff”

To trace the origins of this concept, which already has its own Wikipedia page, we must go back to February 3, 2025. On that day, Andrej Karpathy, a founding member of the OpenAI research team and former director of AI at Tesla, described a laid-back method for developing his “Sunday projects” in a tweet from his personal account. Essentially, it involves verbally instructing Cursor, a code editor based on Claude’s Sonnet models, to fully create an app or website. Then, let it fix the bugs and accept all its suggestions, “going with the flow”. “It’s not really coding,” he tempered. “I see stuff, I say stuff, I launch stuff, and I copy-paste stuff, and it sort of works.” Two years earlier, in January 2023, he had already hinted at this by stating on X that English had become “the most fashionable programming language.”

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

Between Automation, Intuition, and Loss of Control

On February 3, 2025, Andrej Kaparthy painted a picture of a burgeoning reality: delegating the design of a prototype to an LLM, even if it means losing grasp of its architecture. Between acknowledged laziness and creative exploration, this approach has spread within insider circles, propelled by a new generation of “AI first” tools. Platforms like Bolt, Lovable, Replit, and Cursor—unlike pioneers such as GitHub Copilot, which function more as assistants requiring specific skills—place AI at the heart of the creative process. They promise the ability to produce a perfectly adequate and functional program, provided one can articulate their intentions clearly and master the art of the prompt. This differentiates vibe coding quite distinctly from using generative AI merely as an assistant, though the two are often confused. “If an LLM wrote every line of your code, but you reviewed, tested, and understood it thoroughly, to me, that’s not vibe coding—it’s just using an LLM as an assistant,” confirms a developer interviewed by Ars Technica.

Nevertheless, the concept—which initially sparked the same wonder as one’s first encounter with ChatGPT—is quite appealing. And that’s likely why it has quickly “gained traction,” as they say in the industry. The media coverage of astonishing projects, such as a flight simulator developed supposedly with Cursor in three hours by Dutch programmer Pieters Levels, has also helped.

Ok it’s done, you can play it at

https://t.co/6TyHKaj8lb

I’ve never ever made a game before and just made my own flight simulator 100% with Cursor in I’d say 3 hours by just telling it what I wanted

It didn’t go 100% smooth ofc, but 80% yes, a few times I had to go back to… https://t.co/pSQL37c8zf pic.twitter.com/hEdO7O6pPz

— @levelsio (@levelsio) February 22, 2025

“Vibe Coding Might Resemble Witchcraft”

The operation of platforms like Bolt, Cursor, and Lovable is quite similar: the user describes what they want using a prompt, and the tool takes care of generating the program’s architecture, selecting the most suitable resources or languages. Then, the user can fine-tune the result by entering additional queries to fix bugs or edit the visual appearance and positioning of elements, or ask for explanations in an integrated chat. Depending on the project’s complexity, the generation process can take anywhere from a few minutes to several hours. This formula has been the key to the success of startups behind these tools. Anysphere, the company behind Cursor, is already valued at over 9 billion dollars, reports the Financial Times, while the Swedish Lovable, boasting 500,000 users according to TechCrunch, has been raising funds actively in recent months.

“For someone who doesn’t code, vibe coding might seem like magic,” observes Kevin Roose, a tech reporter at the New York Times. The journalist, who has used these tools to create programs capable of determining whether a piece of furniture could fit in the trunk of his car or help prepare his son’s lunch, nonetheless reminds us that vibe coding “still benefits from human supervision, or at least from their being nearby. And it’s probably better suited for personal projects rather than critical tasks.”

Revolution in Progress, or Hollow Marketing Promise?

At this stage, can we truly assess the potential impact of vibe coding? It would be presumptuous. However, this hasn’t stopped some figures in the California tech scene, always eager to hype up a trend that might inflate their portfolios, from making some predictions. In a video simply titled “Vibe coding is the future,” Garry Tan, CEO of the startup accelerator Y Combinator, which has funded companies like Airbnb and Dropbox, states unequivocally: “This isn’t a fad. It’s not going away. It’s the new way to code. And if you’re not on board, you might just miss the boat.”

On forums and social platforms, the practice raises questions and divides opinions, between those who see it as an ongoing revolution and those who dismiss it as shallow or deceptive. “Vibe coding is essentially AI companies marketing to make you pay $200 a month by making you think it’s awesome,” quips a Reddit user. “It’s like buying a race car kit, paying your drunk uncle who ‘knows a bit about racing’ to assemble it for you, then telling all your friends you built it yourself,” adds another.

What’s up with « vibe coding »?

byu/_Amish_Avenger_ inOutOfTheLoop

A Practice That Reignites the Eternal Debate Over the Role of Developers

Whether ignored or praised, vibe coding has certainly reignited a perennial debate that also affects other sectors, such as journalism—the potential deskilling of developers, especially those at the beginning of their careers, in a context where AI is already commonly used to generate code, debug, or answer technical questions, as shown by a study conducted by Stack Overflow in 2024. While it’s possible to create simple functionalities without technical skills and with a bit of instinct, how far can automation go in the profession? The question probably deserves to be asked. However, if the risk is real, it remains to be seen in a sector where maintaining, optimizing, and making the right technical choices primarily depend on understanding what one is doing. As in most professions quickly labeled as threatened by AI, this expertise is difficult to replace.

Rather than being sold as a revolution or presented as the future of programming, vibe coding should be seen for what it is, at least for now: a way to explore ideas or create prototypes, sometimes absurd ones, by “asking the silliest things,” as Andrej Kaparthy suggested in his tweet-turned-manifesto. It’s a way to open doors, for both novices and experts, provided you know what you’re looking for.