What is vibe coding?

In plain English

Vibe coding is a loose, informal way of making software with AI assistance. Rather than writing code line by line, you describe the thing you want — “a page where people sign up for my running club, with a confirmation email” — let the model generate the code, run it, see what breaks, and describe the fix. You steer by results and feel rather than by reading the code.

The phrase was coined in early 2025 by AI researcher Andrej Karpathy, and it caught on because it captures something real: for small projects, people with little or no programming background can now produce working software in an afternoon.

Why it matters

It lowers the barrier to building tools dramatically. A teacher can make a marking tracker, a shopkeeper a stock sheet, a researcher a quick data visualisation — things that previously meant hiring a developer or abandoning the idea. It also changes what junior developers spend time on, moving effort from typing code to specifying, testing and reviewing it.

An example

A cycling club secretary describes a simple ride-booking page to an AI coding tool, tries it, notices members can book the same slot twice, says so in plain English, and gets a fixed version. At no point do they read the underlying code.

What to watch out for

Code you have not understood is code you cannot safely maintain or secure. Vibe-coded projects commonly carry hidden problems: weak handling of personal data, missing checks on user input, dependencies that go stale, and bugs that only appear under real-world load. That is tolerable for a personal tool and risky for anything handling customer details, payments or anything legally regulated. Treat it as a fast way to prototype, and get a qualified human review before anything goes live.