The world of web development is constantly evolving, and a new trend, dubbed vibe coding, has been exciting the tech community since 2025.
Popularized by Andrej Karpathy, former OpenAI researcher and an iconic figure in artificial intelligence, this concept redefines how software is designed and created. But what is vibe coding, where does it come from, and why is it generating so much enthusiasm and debate?
This article, written by the Yiaho team, explores the definition, history, and implications of this revolutionary approach.
What is vibe coding in AI?
Vibe coding, or “intuitive programming,” is a software development method that relies on artificial intelligence, specifically large language models (LLMs) like GPT-4, Claude, or Gemini, to generate code from natural language descriptions.
Rather than manually writing each line of code, the developer (or even a non-expert) describes their idea or objective in simple terms, such as “create a web application with a registration form” or “develop a game where a character avoids obstacles.”
The AI then translates this intention into functional code.
An intuitive method
The essence of vibe coding lies in its intuitive and fluid approach. Unlike traditional programming, which requires mastery of languages, algorithms, and data structures, vibe coding emphasizes creativity and experimentation. This approach can be found with our online code generator on Yiaho, for example.
The term, coined by Andrej Karpathy in February 2025, reflects this philosophy: “go with the vibes, embrace the exponentials, and forget code even exists.” In practice, this means the user guides the AI through instructions, tests the generated code, and iterates until the desired result is obtained, often without fully understanding the technical workings of the produced code.
This approach is also distinguished by its conversational nature. Thanks to tools like Cursor, Bolt, or Replit, users can interact with AI via text or even voice commands, as Karpathy illustrated with his use of SuperWhisper to request simple modifications, such as “reduce sidebar margin by half.” The result? A development experience that feels less like meticulous work and more like a creative collaboration with a machine.
The history of vibe coding: an AI-driven revolution
The beginnings: the rise of generative AI tools
Vibe coding didn’t come out of nowhere. It is part of a broader evolution of generative artificial intelligence, which has transformed many fields, from natural language processing to image and video creation. As early as 2021, tools like GitHub Copilot, based on OpenAI’s Codex model, introduced a first wave of programming assistance by suggesting real-time code completions.
These tools remained assistants for experienced programmers, requiring technical understanding to validate and adjust suggestions.
With the advancements in LLMs, the capabilities of AI tools have expanded.
In 2023, Karpathy himself predicted that “the hottest new programming language is English,” emphasizing that AI models were becoming powerful enough to interpret natural language instructions and produce complex code. This vision took shape with the emergence of tools like Cursor, Lovable.dev, or Bolt, which allow generating entire applications—user interface, backend logic, and database—from simple text descriptions.
The birth of the term: Andrej Karpathy and the 2025 buzz
It was in February 2025 that the term “vibe coding” was officially introduced by Andrej Karpathy in a post on X: “There’s a new way to code that I’m calling vibe coding, where you completely go with the vibes, embrace the exponentials, and forget code even exists.” This message, accompanied by concrete examples such as the creation of an application prototype named MenuGen, ignited Silicon Valley and beyond.
In March 2025, the term was even integrated into the Merriam-Webster dictionary as a “trending slang,” a sign of its rapid adoption.
Vibe coding benefited from a favorable technological context. Platforms like Replit or v0 (Vercel) simplified the creation of full-stack applications in minutes, while editors like Cursor, based on Visual Studio Code, integrated AI agents, as on Yiaho, capable of dialoguing with users to generate, correct, and optimize code.
In parallel, impressive statistics reinforced the enthusiasm: in March 2025, Y Combinator reported that 25% of startups in its winter cohort used codebases 95% generated by AI, demonstrating the impact of vibe coding even among skilled developers.
The strengths and limitations of vibe coding
Democratizing development
Vibe coding has shattered the entry barriers to software development. Now, non-technical profiles—designers, entrepreneurs, marketers—can create functional prototypes without years of learning. For example, New York Times journalist Kevin Roose, a non-programmer, used vibe coding to develop an app called LunchBox Buddy, capable of analyzing his refrigerator’s contents to suggest meals. This ability to transform an idea into a product in a few hours is a revolution for independent creators and startups in the prototyping phase.
For experienced developers, vibe coding acts as a “force multiplier,” freeing up time to focus on overall architecture, strategy, or creativity, rather than repetitive tasks like bug fixing or writing boilerplate code.
Tools like Cursor Workspace or Temps Labs even allow integrating AI agents directly into the development environment, transforming the IDE into an active partner. The use of AI among developers is also growing, despite much distrust.
The challenges: quality, security, and responsibility
However, vibe coding is not without risks.
A key characteristic of this approach is that users often accept the code without fully understanding how it works, which can be problematic in professional contexts. As researcher Simon Willison pointed out: “If you’re reviewing, testing, and understanding every line of AI-generated code, that’s not vibe coding, that’s using AI as a typing assistant.” This distinction highlights a major risk: bugs, security flaws, or inefficiencies in the generated code can go unnoticed, especially if the user lacks technical expertise.
For example, code generated for a database might seem functional but ignore essential practices like password hashing and salting, exposing users to vulnerabilities. Furthermore, LLMs can produce “hallucinations,” erroneous features or behaviors, and AI biases, and struggle to manage complex projects involving multiple files or poorly documented libraries. In a professional context, where maintainability and security are crucial, vibe coding is often relegated to prototyping projects or “disposable weekend projects,” as Karpathy initially envisioned.
Finally, ethical and legal questions arise. Who is responsible if AI-generated code causes a security flaw or data loss? The user, the tool, or the AI model developer? These issues, still largely unclear, highlight the need for a cautious approach.
Outlook: towards an AI-augmented developer?
Vibe coding does not replace traditional programming, but it redefines the developer’s role.
Some speak of an “augmented developer,” capable of delegating technical tasks to AI while focusing on vision, architecture, and validation. As LLMs gain power and technologies like Multi-Context Processing (MCP) servers allow AI to better understand the overall context of a project, vibe coding could become an essential tool for both professionals and amateurs.
However, experts agree on one point: vibe coding does not eliminate the need to understand the fundamentals of development.
AI is a formidable asset for those who already know how to code, but novices risk producing more bugs than solutions without a solid technical foundation. Thus, vibe coding is an opportunity, but it requires a balance between creative intuition and technical rigor.
Also read on this topic: What is Self-Supervised Learning in AI? Explanations
A turning point in development history?
By leveraging advances in generative AI, it democratizes application creation, lowers entry barriers, and unleashes the creativity of developers, whether novice or expert. This concept embodies a vision where natural language becomes the new programming language, transforming coders into conductors of AI agents.
But this revolution is not without its challenges. The risks of bugs, security flaws, and loss of technical skills remind us that vibe coding is primarily a tool, not a magic solution.
For startups, independent creators, and businesses, it offers a unique opportunity to prototype quickly and innovate. But to fully leverage it, one will need to learn to navigate between intuition and expertise, between “vibes” and rigor.
The future of software development may be being written, and it vibrates to the rhythm of AI! And to discover other definitions, find our AI dictionary.
