The short version
The term was coined by Andrej Karpathy in early 2025. He described a way of building software where you tell an AI what you want, it writes the code, and you guide it through feedback and iteration. You don't need to understand every line. You need to understand what you're trying to build and whether what the AI produced actually works.
This isn't theoretical. People are shipping real products this way. Apps, websites, automations, internal tools. Some of them are making real money. The barrier to building software has dropped dramatically, and vibe coding is a big part of why.
How it works
A typical vibe coding session:
- You describe what you want in plain language. "Build me a form that collects a name, email, and message, and sends it to my email."
- The AI (Claude, Cursor, Replit Agent, or similar) generates the code.
- You run it and test it. Does the form appear? Does the email arrive?
- If something is wrong, you describe the problem. "The email isn't sending. Here's the error message." The AI fixes it.
- You iterate until it works.
The tools that support this:
- Claude Code and Cursor work inside your codebase, reading and writing files directly
- Replit Agent and Bolt create entire projects from a description
- v0 (by Vercel) generates UI components from prompts
- Lovable and Windsurf focus on full-app generation
What makes vibe coding different from "just using ChatGPT to write code" is the feedback loop. The AI has context about your whole project, not just a single question. It can read error messages, check what's already built, and make changes across multiple files.
Why it matters
Vibe coding doesn't replace professional software engineering. Complex systems, performance-critical code, and security-sensitive applications still need people who understand what's happening under the hood. But for building MVPs, internal tools, personal projects, and small businesses, it's a real shift. People who couldn't build software a year ago are shipping products today.