The Basics: What is Vibecoding, and Can I Do It Too?
3 minute read
The NYT columnist Kevin Roose wanted to test the limits of what AI could do. The first task: his son’s school lunch. He asked AI to build him an app that would suggest healthy, peanut-free lunch options – all by analysing a photo of his fridge.
The AI got to work. It mapped out the task, built a basic interface, chose an image-recognition tool and began writing code. If the AI needed him to make a decision – does the app need to disclose nutritional facts? – it asked. When it got stuck, it reverted to an old version and tried again.
Ten minutes later, “LunchBox Buddy” app was live and ready to use. Roose hadn’t written a single line of code.
Welcome to vibecoding.
Never Miss A Story
What is vibecoding?
For decades, coding meant learning a new language – Java, Python, C++. Vibecoding flips that; instead of writing code, you describe what you want – in plain language – and AI generates the code for you.
OpenAI co-founder Andrej Karpathy popularised the term in early 2025, and it quickly spread from hobbyists to Silicon Valley executives.
Instead of writing code, you describe what you want – in plain language – and AI generates the code for you.
Why now?
The casual, approachable term makes vibecoding feel accessible. As Yale management professor Kyle Jensen notes, “‘A.I.-assisted software development’ doesn’t roll off the tongue”.
People can now build apps that suggest how many layers they should wear today, extremely specific workout plans, teaching budgeting to kids, a calendar availability sharing chrome extension – all, theoretically, without formal coding training.
What if it goes wrong?
AI systems can generate broken code or wander off in unexpected directions, so human guidance is often needed. People with a programming background often find it easier to spot problems and guide the process back on track.
Simon Last, founder of software start-up Notion, compares it to managing interns. While journalist Lauren Goode more critically said she felt like “a responsible babysitter for code.”
“‘A.I.-assisted software development’ doesn’t roll off the tongue.”
Kyle Jensen
How are businesses using it?
Many argue it’s just a rudimentary tool for one. “These aren’t the kinds of tools a big tech company would build. Their features are limited and some of them only sort of work”, says Kevin.
But its broader shift is already visible. At Google, Sundar Pichai has said that AI-generated code now accounts for more than a quarter of new code deployed. The principle feels similar: people prompt, AI generates.
As with much of the conversation around AI, vibecoding raises familiar questions: around quality control, intellectual property, and what the future workforce might look like. But it is unlikely to replace traditional programming altogether. Bugs will still appear, and people will always need to be there to fix them.
We might be “vibing” our way through a human-AI relationship that’s still in flux – perhaps that’s exactly what makes the term feel so right.
