ARTIFICIAL INTELLIGENCE

“Vibe coding is an AI-driven software development approach where users describe desired app features in natural language (the “vibe”), and a Large Language Model (LLM) generates the functional code.” – Vibe coding

Vibe coding is an AI-assisted software development technique where developers describe project goals or features in natural language prompts to a large language model (LLM), which generates the source code; the developer then evaluates functionality through testing and iteration without reviewing, editing, or fully understanding the code itself.^1,2

This approach, distinct from traditional AI pair programming or code assistants, emphasises “giving in to the vibes” by focusing on outcomes, rapid prototyping, and conversational refinement rather than code structure or correctness.^1,3 Developers act as prompters, guides, testers, and refiners, shifting from manual implementation to high-level direction—e.g., instructing an LLM to “create a user login form” for instant code generation.² It operates in two levels: a tight iterative loop for refining specific code via feedback, and a broader lifecycle from concept to deployed app.²

Key characteristics include:

• Natural language as input: Builds on the idea that “the hottest new programming language is English,” bypassing syntax knowledge.¹
• No code inspection: Accepting AI output blindly, verified only by execution results—programmer Simon Willison notes that reviewing code makes it mere “LLM as typing assistant,” not true vibe coding.¹
• Applications: Ideal for prototypes (e.g., Andrej Karpathy’s MenuGen), proofs-of-concept, experimentation, and automating repetitive tasks; less suited for production without added review.^1,3
• Comparisons to traditional coding:

Tools supporting vibe coding include Google AI Studio for prompt-to-app prototyping, Firebase Studio for app blueprints, Gemini Code Assist for IDE integration, GitHub Copilot, and Microsoft offerings—lowering barriers for non-experts while boosting pro efficiency.^2,3 Critics highlight risks like unmaintainable code or security issues in production, stressing the need for human oversight.^3,6

Best related strategy theorist: Andrej Karpathy. Karpathy coined “vibe coding” in February 2025 via a widely shared post, describing it as “fully giv[ing] in to the vibes, embrac[ing] exponentials, and forget[ting] that the code even exists”—exemplified by his MenuGen prototype, built entirely via LLM prompts with natural language feedback.¹ This built on his 2023 claim that English supplants programming languages due to LLM prowess.¹

Born in 1986 in Bratislava, Czechoslovakia (now Slovakia), Karpathy earned a BSc in Physics and Computer Science from University of British Columbia (2009), followed by an MSc (2011) and PhD (2015) in Computer Science from University of Toronto under Geoffrey Hinton, a neural networks pioneer. His doctoral work advanced recurrent neural networks (RNNs) for sequence modelling, including char-RNN for text generation.¹ Post-PhD, he was a research scientist at Stanford (2015), then Director of AI at Tesla (2017–2022), leading Autopilot vision—scaling ConvNets to massive video data for self-driving cars. In 2023, he co-founded OpenAI’s Supercluster team for GPT training infrastructure before departing in 2024 to launch Eureka Labs (AI education) and advise AI firms.^1,3 Karpathy’s career embodies scaling AI paradigms, making vibe coding a logical evolution: from low-level models to natural language commanding complex software, democratising development while embracing AI’s “exponentials.”^1,2,3