History of Vibe Coding

In February 2025, Andrej Karpathy -- a co-founder of OpenAI and former head of AI at Tesla -- posted a tweet describing a new way he'd been building software: talking to an AI in plain English, accepting its suggestions, and barely looking at the actual code. He called it 'vibe coding,' and the term spread across the programming world within days.

- Where the term 'vibe coding' came from and what it originally meant - How coding with AI has changed since the 1950s - The difference between traditional coding and vibe coding - Why vibe coding is controversial among professional programmers

In the 1950s, programmers wrote code on punch cards -- stiff paper cards with holes in specific spots that a computer could read. A single mistake meant re-punching the whole card. By the 1980s, programmers typed text commands directly into a computer using languages like C. In the 2010s, tools called autocomplete started guessing the next few characters you might type. Then in 2021, GitHub released an AI tool called Copilot that could write whole lines of code based on what you were doing. Vibe coding, named in 2025, is the next big jump: instead of writing code yourself, you describe what you want in plain sentences, and an AI writes the entire program.

Karpathy described vibe coding as 'giving in to the vibes' -- you tell the AI things like 'make the button bigger' or 'add a login page,' and when something breaks, you just paste the error message back to the AI instead of debugging it line by line yourself. This is very different from traditional coding, where a programmer studies every line, understands exactly why it works, and can explain each decision. Vibe coding trades that deep understanding for speed: someone with no programming background can build a working app in an afternoon.

By late 2025, tools like Claude Code, Cursor, and Replit's Agent had made vibe coding common not just for hobby projects but for real startups. Many companies now blend both approaches: they vibe code a first draft quickly, then have an experienced engineer review and fix the AI's work before it goes live to real users. This blended approach is often called 'AI-assisted development' rather than pure vibe coding, because a human still checks the details.