Science Editor Mony Aramalla examines how “vibe coding” is transforming software development, from AI-generated apps to the new challenges developers face in the age of autonomous coding agents.
Exactly a year ago, the word ‘vibe coding’ was coined by AI researcher and OpenAI co-founder Andrej Karpathy to describe a computer programming workflow that is predominantly reliant on AI agents rather than manual coding. The name itself comes from giving in to the “vibes” while using everyday language to prompt Large Language Models (LLMs) to build apps, sites, and startups, reducing the need to read or understand the code generated. By early 2026, industry surveys suggest that over 90% of developers use some form of AI-assisted coding daily.
The ease that comes with simply generating lines of code is that there’s now what we call a “Quality Tax.” Without the need to write the code functions themselves, senior developers increasingly spend their time reviewing, debugging, and evaluating the code generated by these autonomous agents.
So, has programming disappeared?
The first wave of AI coding came about in 2023 and 2024 in the form of generated snippets. Today’s tools operate more like autonomous machines that embed directly into the developer’s environment.
Lovable and v0: For decades, developers considered designers enemies because of how difficult it was to translate mockups into functioning software. Tools like Lovable and v0 have eliminated that layer. A designer can upload a screenshot of a sketch or a Figma prototype and receive a fully functional frontend design in seconds. Non-designers and engineers can now produce functional interfaces. Entire MVPs can be made in hours rather than weeks.
Windsurf and Cursor: How developers interact with their codebases is important. That’s where Windsurf and Cursor come in. Their environments act less like text editors and more like reasoning systems. Instead of asking the AI to fix a single bug, developers can issue instructions:
“Make this page feel more minimalist in colour and design.”
Automatically, the agent finds the relevant components, rewrites the code, and changes dozens of files at the same time. Developers don’t worry about the inputs anymore; they focus on changing the output of the AI agent.
While startups and many senior developers are utilising vibe coding to simplify their tasks, the concept has been beneficial in places that traditionally lacked access to software development.
In a small rural clinic working with limited resources and an ageing medical record system, the clinic needed two simple tools: a Prednisone taper calculator that adjusted steroid dosage schedules and an insulin blood sugar simulator that helped clinicians model patient treatment adjustments.
Large medical device companies gave quotes reaching hundreds of thousands of euros. But one clinician with no formal programming experience experimented with AI development software like Bolt.new and SuperNinja. Using these tools, they were able to build small decision-support tools.
The total cost of building the system? Under $30 in API credits.
This is what vibe coding has created: a democratisation of the app where people closest to real-world problems can finally build their own solutions.
However, the rapid rise of vibe coding has also introduced new challenges.
One of the most discussed issues is skill atrophy. Even Andrej Karpathy himself said that he has begun to lose his ability to write complex code manually. When developers rely so heavily on AI agents, they lose the habit of coding, which raises the uncomfortable question: if developers stop practicing coding, who will be responsible for auditing these AI models?
An example of the hidden risks of vibe coding emerged in a widely discussed developer forum thread in 2026. A startup founder built an application almost entirely through AI prompts. The app worked perfectly during testing, but when the team examined the database, they discovered a serious flaw. The AI stored all user data in a single JSON column. This fulfilled the prompt’s requirements, but the system was extremely difficult to work with. The team had to spend more resources fixing the issue rather than building it correctly.
Despite these concerns, vibe coding is changing what it means to be a developer.
For years, programming skill was defined by how well someone wrote precise code and how effectively they could debug errors. But in an era of agentic development, syntax is less important. Ideas, judgement, and taste are.
In many ways, a senior developer in 2026 is more like a department head managing thousands of digital workers. So, it’s fair to say that programming has not disappeared, but evolved.