"Vibe coding," an emerging software development paradigm where artificial intelligence generates code from natural language prompts, is rapidly gaining traction, with a significant impact on the startup ecosystem. Coined by OpenAI co-founder Andrej Karpathy in February 2025, this approach allows developers and even non-programmers to create applications by focusing on desired outcomes rather than intricate syntax. The concept, often described as using English as the "hottest new programming language," streamlines the development process.
This innovative method involves users describing their project or task to a large language model (LLM), which then produces the necessary code. Unlike traditional AI-assisted coding, "pure" vibe coding often sees developers accepting AI-generated output without extensive manual review or editing, instead focusing on iterative experimentation and refinement. Dan Shipper highlighted this trend, tweeting, "> learn to vibe code like a pro," and linking to an article about the phenomenon.
The adoption of vibe coding is already evident in the tech industry, with Y Combinator reporting in March 2025 that 25% of startups in its Winter 2025 batch had codebases that were 95% AI-generated. This statistic underscores a significant shift towards AI-centric development within nascent companies. The practice is particularly beneficial for rapid prototyping and democratizing app creation, enabling individuals without deep programming knowledge to bring ideas to life quickly.
However, the approach is not without its challenges and critics. While it lowers the barrier to entry and boosts productivity, concerns exist regarding code quality, debugging complexities, and the potential for inconsistent design patterns if the underlying code is not understood or reviewed. Programmer Simon Willison noted that true vibe coding involves accepting AI output without fully understanding it, differentiating it from responsible AI-assisted programming where human review is paramount.
Despite these considerations, platforms like Replit, Cursor AI, and GitHub Copilot are facilitating vibe coding, offering tools that transform natural language into functional code. As the technology evolves, with trends like multimodal programming incorporating voice and visual inputs, vibe coding is set to further reshape how software is built. It represents a fundamental change from manual coding to a more conversational, AI-driven development workflow.