The guy who coined 'vibe coding' predicts it will 'terraform software and alter job descriptions'

Andrej Karpathy, known for coining the term "vibe coding" earlier this year, has shared his insights on how this concept is transforming the software engineering field. With an extensive background that includes five years at Tesla, where he led the Autopilot project and briefly worked on the humanoid robot Optimus, Karpathy's expertise is backed by his experience as a co-founder of OpenAI. Reflecting on the evolution of large language models (LLMs) in his year-end recap on X, Karpathy emphasized that vibe coding democratizes programming. He noted that it opens the door for non-professionals to engage in coding, allowing a broader audience to harness the power of LLMs. According to Karpathy, this shift is not just beneficial for individual users but also advantageous for businesses as they equip their engineers with advanced tools like Cursor, Claude Code, and OpenAI's Codex to enhance productivity. Karpathy described vibe coding as a movement that enables professionals to produce software that might never have been created otherwise. He characterized this new coding paradigm as "free, ephemeral, malleable, and discardable after single use," suggesting a significant departure from traditional coding practices. He humorously reflected on the unexpected popularity of the term he coined, admitting he was initially unaware of its potential impact. The concept allows developers to immerse themselves in creativity, leveraging sophisticated LLMs to simplify the coding process. Yet, the efficiency of vibe coding remains a topic of debate; a recent METR study found that AI coding assistants actually reduced productivity among experienced developers by 19% despite their expectations of increased efficiency. Despite the varying opinions on productivity, the rise of vibe coding is undeniably fostering innovation in tech. For instance, Twitter founder Jack Dorsey successfully vibe-coded a new messaging application this year, illustrating how non-technical individuals can now build and deploy apps in a fraction of the time it once took. Karpathy also shared his enthusiasm for advancements in AI, praising Google's Gemini's Nano Banana image model and recognizing Claude Code as a landmark example of what LLM agents can achieve. Overall, he concluded that 2025 was an exhilarating year for large language models, filled with surprises and breakthroughs.