What is Vibe Coding, Really?

If you’re tuned into the technology space, the term “vibe coding” might mean something to you. Whether it's poorly written AI code or just the act of using AI to write code, everyone seems to have a different understanding of it. The truth is, vibe coding doesn’t mean what people think it means anymore. This article aims to deconstruct this term and its use cases.

What is Vibe Coding?

Andrej Karpathy, a renowned AI researcher and cofounder of OpenAI, coined the term “vibe coding” in a 2025 tweet, and it captures a new era of programming.

At its heart, vibe coding refers to the workflow one can achieve using AI models to create code, and this human-AI interaction allows the creators to “forget that the code even exists.”

The methodology he describes in his tweet refers to the almost effortless process where Large Language Models (LLMs) like Cursor Composer with Sonnet enable him to barely touch the keyboard, relying on voice commands ("I just talk to Composer with SuperWhisper") for even "the dumbest things like 'decrease the padding on the sidebar by half' because I'm too lazy to find it.” For Karpathy, "The code grows beyond my usual comprehension," turning project building into an intuitive flow of "see stuff, say stuff, run stuff, and copy paste stuff, and it mostly works." Kaparthy marks the end of his tweet by claiming, “It’s not too bad for throwaway weekend projects.”

Karpathy’s tweet coining the term “vibe coding”

Programming by Intent has been a long-standing concept in computer science where developers declare what they want the software to do, rather than detailing how to achieve it. This would lead programmers to use specialised languages such as SQL and HTML to focus on the results of what they wanted, or would involve them building complex layers of software to simplify interactions, also known as abstraction layers. Vibe coding brings this concept into a whole new light, as it’s arguably the most accessible and practical manifestation of programming by intent that we’ve seen to date. Vibe coding pushes a new extreme of literally removing the code from the user's direct focus.

Vibe coding consists of an iterative process: users type out what they want to achieve into an LLM, they implement the code, paste errors in if it so occurs, and refine their next prompt depending on their needs. Users literally can type out what they want, and can have functioning software within hours.

Not to mention, the concept of allowing one to “forget the code even exists” is not a disregard to the quality of code, rather, it’s more about achieving a lower cognitive load. By handing off detailed syntax and nuances of coding to an AI, creators can spend mental energy into the creative aspects and the desired outcome itself. Though this is universal in everyone that vibe codes, it’s important to talk about the process of vibe coding for different people.

So, who is vibe coding actually for?

It’s fair to say that Vibe Coding is for everyone, but I would boil it down to two subgroups: non-programmers and programmers. While both embrace the “forget the code even exists” experience, their reasons differ. 

The key is that non-programmers don’t read the code because they don’t understand it, and they don't need to. It’s about direct creation rather than technical understanding.

For those without programming experience, the primary motivation of vibe coding is that it allows them to automate tedious tasks that are specific to them. Whether it’s an application to track expenses or a to-do list, it can yield great results. The fact is, there are many who will not learn programming but still want to build these projects for themselves, which is completely fair. Learning how to code and programming has a steep learning curve, and breaking into such skills requires a large time commitment. Of course, some may find an interest and go on to be skilful software developers, but many will not. 

For the programmers, vibe coding can still bring value, but primarily within low-stakes, personal or even experimental projects. Yes, it can serve the same purpose by automating tasks specific to them, but in this context, it’s more of a “why not let it rip” mindset, driven by the incredibly fast and frictionless nature of vibe coding. Programmers can prototype and tinker rapidly, all while exploring new ideas without worrying about implementation. For programmers they don't read the code because they don't want to or need to. It’s all about tinkering and exploration and high-speed prototyping.  

What Vibe Coding has Become

Due to the dilution of this term, many carry different meanings and interpretations of what vibe coding is. It has gone far astray from its intentional and purposeful definition and has become a buzzword for anything related to AI-generated code. Thus, it’s necessary to talk about misconceptions to ensure miscommunication is avoided. As just mentioned, a main misconception is that AI-written code equates to vibe coding. 

Many tech giants today have AI-generated code in their code base, take, for example, Microsoft, whose code base consists of 20-30% AI-generated code. This widespread use of AI-written code in industry is fundamentally different from what Andrej Karpathy coined as vibe coding.

By definition, vibe coding is unchecked and unmoderated code. The approach where users “forget the code even exists” stands completely contrary to its use case in industry. AI-generated code used in an enterprise undergoes rigorous scrutiny and is meticulously tested and reviewed. Because this code is subject to such human oversight, it can’t be classified as being vibe coded. I believe that for something to be vibe coded, it would be required to be 100% AI-generated with 0 human moderation of the output itself. 

Furthermore, skilled programmers simply do not vibe code in professional roles. They are skilled and adept in their craft, and engaged with the technical intricacies of their work. Hence, the use of AI assistance in this context is simply a part of software development as we know it today. 

Conclusion

Vibe coding shouldn’t carry a negative connotation, it should be recognised as a symbol of how far technology has come. It represents a shift where creating software is no longer reserved for those with deep technical expertise. Instead, it empowers anyone, regardless of coding background, to build tools, automate tasks and bring their ideas to life through natural interaction with AI.

Vibe coding is not just any use of AI in programming, it’s a unique workflow where the code fades into the background, and the focus shifts entirely to intent and outcome. It’s fast, frictionless, and unmoderated. Precisely making it so powerful for side projects, personal tools and individual creativity.

It represents a glimpse into a future where building with code becomes as natural as thinking out loud.

References

Chauhan, V. (2025). What is Vibe Coding and How Does it Work? A Comprehensive Guide. [online] Prismetric. Available at: https://www.prismetric.com/what-is-vibe-coding/ [Accessed 31 Jul. 2025].

Roberts, W. (2025). Vibe coding: What it is, and why you should give it a miss. [online] Digital Trends. Available at: https://www.digitaltrends.com/computing/what-is-vibe-coding-and-is-it-useful/ [Accessed 31 Jul. 2025].

Willison, S. (2025). Two publishers and three authors fail to understand what ‘vibe coding’ means. [online] Simon Willison’s Weblog. Available at: https://simonwillison.net/2025/May/1/not-vibe-coding/ [Accessed 31 Jul. 2025].