Vibe coding
What is Vibe coding?
Building software by describing what you want and accepting what the AI writes, without reading or reviewing the code. Andrej Karpathy coined the term in early 2025 for low-stakes work: prototypes, throwaway tools, weekend projects. The term is narrow. When a professional reviews, tests and stands behind the output, that is AI-assisted development, not vibe coding.
Why it matters
The mode rests on a single trade: understanding for speed. On disposable work that is a fair deal, since the aim is to see something run, not to own it, and reading code you will delete is wasted effort. The bargain sours when the output stops being disposable. Software no one has read cannot be safely maintained, extended or secured, and the understanding skipped has to be rebuilt later or done without. So the real hazard is seldom the mode itself. It is scope creep: the weekend project that gets demoed, then leaned on, until something no one can explain is running where it counts.
Where does the term come from?
Andrej Karpathy named vibe coding in a post on X in February 2025. He described giving in to the vibes and forgetting that the code even exists: he accepted every change the AI proposed and no longer read the diffs. He called it not too bad for throwaway weekend projects. Within a year the phrase had reached general usage: Collins Dictionary named vibe coding its Word of the Year 2025 in November 2025, defining it as the use of AI prompted by natural language to write computer code. That dictionary sense is broader than Karpathy’s original one. This glossary uses the narrower meaning, in which the code goes unreviewed.
Is all AI-assisted coding vibe coding?
No. Vibe coding is not a catch-all for writing software with AI. Simon Willison defines it as building software with a language model without reviewing the code it writes. If a developer reviews the code, tests it and can explain how it works, Willison says, that is software development, however much of it the model wrote. Both ways of working fall under AI-assisted software development, the umbrella term for building software with AI. Vibe coding is its narrow, unreviewed end. IBM draws the same line from the other side: AI-generated code still needs code review, testing, dependency checks and a security assessment before it goes into production.
When does vibe coding fit, and when does it not?
Vibe coding fits when little can go wrong: a prototype to test an idea, a personal script, a throwaway demo, a weekend experiment. Willison’s test is the harm the code could cause if it has bugs or security holes. The lower the possible harm, the more reasonable it is to skip reading the code.
Vibe coding does not fit software that other people depend on. In May 2026 Willison wrote that vibe coding software for other people is grossly irresponsible, because it handles their information and they are the ones hurt by its bugs. The same applies to anything that stores personal data, moves money, connects to other systems or has to be maintained for years. There the missing review is exactly the risk, and a working demo says nothing about how the code behaves under real load or real attacks.
In practice
Someone spins up an internal dashboard over a weekend by describing it and taking whatever comes back. It works, so a team adopts it, and soon it sits inside a real workflow. Six weeks later it breaks, and even the person who built it cannot say why each fix makes things worse. As a throwaway it was fine. The trouble was that its move into production was never a decision anyone made.
Frequently asked questions
What is vibe coding?
Vibe coding is building software by describing what you want to an AI and accepting the code it writes without reading or reviewing it. The term is narrow: it describes a way of working, not every use of AI in programming.
Who coined the term vibe coding?
Andrej Karpathy coined the term in a post on X in February 2025 and meant it for throwaway weekend projects. By November of the same year, Collins Dictionary had chosen vibe coding as its Word of the Year 2025.
Is vibe coding safe for production software?
Not without review. Code that nobody has reviewed cannot be relied on for security, maintenance or data protection. Production software needs review, testing and a security check, whether a person or an AI wrote the code.
What is the difference between vibe coding and AI-assisted software development?
Review. Vibe coding is one narrow way of working within AI-assisted software development: the end where nobody reads the AI’s code. Everywhere else in AI-assisted software development, a developer reviews, tests and takes responsibility for every change, even when the AI wrote most of it.
Sources
- Andrej Karpathy: the original vibe coding post on X (February 2025) – the origin of the term and its intended use for throwaway weekend projects.
- Simon Willison: Not all AI-assisted programming is vibe coding (but vibe coding rocks) (19 March 2025) – definition, the low-stakes test and the line to software development.
- Simon Willison: Vibe coding and agentic engineering are getting closer than I’d like (6 May 2026) – vibe coding software for other people is irresponsible.
- IBM Think: What is vibe coding? – generated code still needs review, testing, dependency checks and security assessment before production.
- Collins Dictionary: Word of the Year 2025 (6 November 2025) – vibe coding as Word of the Year and its dictionary definition.
- Wikipedia: Vibe coding – overview of the term, its reception and criticism.