What isvibe coding?
Published
Last updated
Our take
Vibe coding is building software by describing what you want to an AI tool (like Cursor, Lovable, or Bolt) in plain language and accepting the code it writes, often without reviewing it. Coined by Andrej Karpathy in February 2025, it's fast and ideal for prototypes, but risky for production without expert review.
This answer covers
- Vibe coding
- AI-generated code
- AI-assisted development
Vibe coding describes a way of working where you prompt an AI (Lovable, Cursor, Bolt, Replit, and similar) in everyday language and accept the code it generates, often without reading it line by line. The name comes from a February 2025 remark by Andrej Karpathy, former director of AI at Tesla, about “giving in to the vibes” and forgetting the code even exists.
Why it took off
It collapses the distance between an idea and a working screen. A non-technical founder can go from a sentence to a clickable app in an afternoon, with no team and no setup. For validating a concept, building a demo, or testing a prototype with users, that speed is genuine and useful.
Where it holds up, and where it doesn’t
Vibe coding is dependable for throwaway prototypes, internal tools, and early validation. It becomes risky the moment real users, real data, or real money are involved. The AI optimizes for “does it run,” not “is it secure, will it scale, can anyone maintain it.” Common gaps are missing authentication, no tests, exposed API keys, and structure no other developer can safely change.
Vibe coding vs AI-assisted development
They are not the same. In vibe coding, the AI builds and the human mostly trusts. In AI-assisted development, an experienced engineer directs the AI and reviews everything it produces. The first is for speed; the second is for shipping software people depend on.
Key takeaways
- Vibe coding means generating software by prompting AI in natural language, not by handwriting code.
- The term was coined by Andrej Karpathy in February 2025.
- It is excellent for prototypes and validation but risky for production without review.
- Typical gaps: security, testing, scalability, and long-term maintainability.
- It differs from AI-assisted development, where a senior engineer stays in control.
Was this answer helpful?
