Vibe coding is the current hype in software development. Can everyone code today with the help of AI? It’s not that simple.
Vibe coding seems well on its way to becoming the IT word of the year. The AI craze is also finding its way into the world of software development. The wide range of artificial tools makes coding seem like child’s play: you describe what you want, and moments later you have ready-made software.
This sounds like every developer’s dream. You save yourself hours by not having to write code manually anymore. Vibe coding also makes coding accessible to non-technical profiles. Can every professional developer retire early?
From Prompt to Code
The term vibe coding has only recently emerged. Andrej Karpathy, one of the co-founders of OpenAI, can be credited with coining the term. Karpathy describes vibe coding as follows: “I just see things, say things, do things and copy and paste things, and usually it works”.
Coding with the help of AI is not new: Github Copilot has existed since 2021, which is already an eternity ago in the AI world. The idea is that as a developer, you no longer write code, but describe it. The AI coding assistant creates the code based on a textual description.
This means that as a programmer, you no longer need to specialize in a specific programming language to write code in it. “But vibe coding is not necessarily for non-developers. There are also tools tailored for professionals,” says Menno Odijk, Field CTO at Siemens subsidiary Mendix.
I just see things, say things, do things and copy and paste things, and usually it works.
Andrej Karapthy, co-founder OpenAI on vibe coding
Twenty Percent
Odijk is not opposed to such tools, but sees limitations in applying them in a professional development environment. “With vibe coding, you can quickly create a prototype that looks like what you’re looking for. Many experiments are happening, but they run into limitations. The last twenty percent often costs eighty percent of your time”.
He clarifies: “Development of business software follows a standardized building process with blocks, frameworks, and standards on how the software should be built. You quickly run into issues with vibe coding there. The time you gain is immediately lost again when getting the code ‘production-ready’. That’s the last twenty percent I’m talking about”.
This is confirmed by Luigi van den Borne, web developer and co-founder of the Belgian company We Are Knights. “From experience, I find AI useful for small, well-defined questions. For large projects, such as (re)writing code or code reviews, it’s less reliable. The biggest limitation of vibe coding, I find, is that there’s often no clear line in the answers: it sometimes feels like a collection of pieces that have been taken from the internet”.
Haste Makes Waste
“Making models work within defined structures is one of the big themes of the moment. AI is creative, but in professional development, you can’t be too creative with standards. It’s difficult to look ahead, but I don’t see us building business applications solely with prompts anytime soon. Quality shouldn’t come at the cost of speed,” Odijk continues.
That doesn’t mean that every developer needs to be fluent in multiple programming languages. Odijk’s employer Mendix is heavily invested in low code: developing applications with as few lines of code as possible. According to him, this can be perfectly reconciled with AI-driven coding.
“AI tools give you code, but people who have no knowledge of development will still not be able to read that code. Low-code then bridges the gap because you work with visual models instead of code. This gives you a bit less freedom because you have to stay within building blocks, but we strongly believe it’s the better way to develop software”.
AI at the Front End
There are situations where vibe coding can add value, Odijk acknowledges. “It can be useful for designing the front end of the application because you can describe in words what you want to see. The tools can design something that looks good. For an individual user, it might be possible to design a complete application with prompts”.
There’s no shortage of creative examples. Today, there are already quite a few applications that have been fully developed with vibe coding, from ‘SEO calculators’ to a free DocuSign alternative. But for every project that succeeds, there are countless that fail.
“The majority of AI projects never make it to production,” says Odijk. “Knowledge and data are usually the obstacles. Within many organizations, there’s still insufficient knowledge about how to approach AI projects from problem to solution. LLMs don’t do everything. This requires expertise and the right tooling that you need to build or hire. The technology is evolving very rapidly”.
“AI is not magical. Structuring data is crucial because it forms the raw material for AI. We’ve gotten used to AI, now we need to make sure we make data suitable so that the tools can do something meaningful with it. This aspect is still underestimated”.
Vibe coding sometimes feels like a collection of pieces that have been taken from somewhere on the internet.
Luigi van den Borne, co-founder We Are Knights
Enthusiastic Intern
The conversation with Odijk suddenly takes a philosophical turn. “Vibe coding is about how you develop software, but what is software? That’s changing as well. AI is now also in the code to autonomously perform tasks. This requires different skills than classic software development. As a developer, you don’t always know anymore what your software will do”.
As software changes, so does the life of the developer. Odijk: “The basic skills will remain relevant. But in a few years, it may no longer be necessary to specialize in specific programming languages. The focus shifts to prompt-based writing and maintaining quality and guardrails. A programmer becomes more of a controller”.
“The profession won’t disappear, I’m convinced of that, but the task package changes. Understanding what the client wants and specifying that becomes a more important skill than writing code. AI tools will be able to quickly replace a junior programmer, but experienced developers will continue to play an important role. But how will you get seniors if you no longer need juniors?” Odijk ends with an open question.
read also
Writing or Checking Code? “Remain an Expert in What You Ask AI to Do”
Van den Borne: “The core remains that as a developer, you must be able to read, understand, and critically evaluate the code. The role of the programmer is gradually shifting from writing everything yourself to more directing, reviewing, and fine-tuning. AI is increasingly taking over repetitive and everyday tasks, allowing me to spend more time on more complex projects”.
“Compared to five years ago, I notice that finding bugs is much smoother: what used to sometimes take hours is now often solved within minutes. It feels a bit like having a good intern next to you who is always up to date with the latest techniques and enthusiastically helps you, but occasionally misses the mark,” concludes van den Borne.