- Type
- AI-assisted development practice
- Coined by
- Andrej Karpathy
- Coined
- February 2025
- Core idea
- Describe intent, AI writes code
- Recognition
- Collins Word of the Year 2025
- Key risk
- Review, security, maintainability
- Type
- AI-assisted development practice
- Coined by
- Andrej Karpathy
- Coined
- February 2025
- Core idea
- Describe intent, AI writes code
- Recognition
- Collins Word of the Year 2025
- Key risk
- Review, security, maintainability
Vibe coding is a software development practice, assisted by artificial intelligence, in which a developer describes a desired program or task in natural language and a large language model generates the corresponding source code. The term was coined in February 2025 by the computer scientist Andrej Karpathy, a co-founder of OpenAI and former AI leader at Tesla. In its purest form, vibe coding involves the programmer guiding, testing, and giving feedback on AI-generated code rather than writing it line by line, sometimes accepting output without reading it in detail and steering the result through further prompts.
Origin and definition
The concept extends an earlier observation by Karpathy that, given the capabilities of modern language models, the hottest new programming language is English, meaning that people could increasingly command computers through natural language rather than formal code. Vibe coding names the workflow that follows from this idea: the human expresses intent and overall direction, the model produces and revises the implementation, and the human evaluates the behaviour of the result. Karpathy described the experience as leaning into the vibes, accepting the model's suggestions, and iterating quickly rather than scrutinising every detail.
The term spread rapidly. Merriam-Webster listed it in March 2025 as a slang and trending expression, and Collins English Dictionary named it the Word of the Year for 2025. By mid-2025, reporting indicated that the practice was moving from hobbyist experimentation into professional and commercial software development.
How it differs from general AI-assisted coding
Not all AI-assisted programming is vibe coding. Tools that suggest completions or that a developer uses while still reading and understanding every change represent a more conventional, review-heavy workflow. Vibe coding specifically describes the looser style in which the developer relies primarily on results and follow-up prompts to drive changes, paying less attention to the underlying code. Commentators have noted that this distinction matters: the same tools can be used carefully or casually, and only the casual, results-first mode is properly called vibe coding.
A common framing describes a spectrum or set of levels of AI involvement in development, from manual coding with occasional assistance, through heavy code-completion, to fully prompt-driven generation in which the human rarely inspects the code at all. Vibe coding sits at the high-automation end of this spectrum.
Benefits and adoption
The appeal of vibe coding is speed and accessibility. It lowers the barrier to building working software, allowing people with limited programming experience to create prototypes, scripts, and small applications, and it lets experienced developers move quickly through routine tasks. This has made it popular for rapid prototyping, internal tools, and exploratory projects, and a range of products marketed as AI coding assistants and agents support the workflow.
Criticisms and risks
Critics caution that vibe coding carries real risks when applied carelessly. Accepting generated code without thorough review can introduce security vulnerabilities, subtle bugs, and code that is difficult to maintain, because no human fully understands how it works. Concerns about accountability arise when no one can explain or confidently modify the resulting system. Practitioners and researchers, including those who otherwise welcome AI assistance, have stressed that vibe coding is well suited to low-stakes and throwaway projects but requires far more caution for production systems, where understanding, testing, and review remain essential. As a result, professional adoption tends to pair AI generation with disciplined practices such as automated testing, code review, and security scanning.
Vibe coding has clear implications for Malaysia's growing technology workforce and startup scene. By lowering the barrier to building software, AI-assisted development can help small and medium enterprises and early-stage founders create digital products with leaner teams, supporting the entrepreneurship and digital-economy goals of the MyDigital Blueprint and agencies such as MDEC.
For the talent ecosystem, the rise of vibe coding shifts the skills that matter. Programmes supported by HRD Corp and universities increasingly emphasise prompt fluency, system design, testing, and the ability to evaluate AI-generated code, rather than rote syntax alone. This is relevant to Malaysia's ambition to expand its pool of digital and AI talent, although educators caution that foundational understanding remains necessary to use these tools safely.
The security concerns associated with vibe coding intersect with the mandate of NACSA (the National Cyber Security Agency) and CyberSecurity Malaysia. Code generated without careful review can introduce vulnerabilities, which is a particular worry for applications handling personal data under the Personal Data Protection Act (PDPA) or operating in regulated sectors such as banking and finance overseen by Bank Negara Malaysia (BNM). Malaysian organisations adopting AI-assisted development are therefore encouraged to combine the speed of vibe coding with disciplined review, testing, and security practices.