Vibe Coding: 25% of Startups Use AI for 95% of Code

Vibe coding, a novel approach to software development, has emerged as a significant trend in the tech industry. This method involves using AI to generate code based on natural language instructions, allowing developers to focus on high-level concepts rather than the intricacies of coding syntax. Andrej Karpathy, a co-founder of OpenAI and former AI director at Tesla, popularized the term "vibe coding" on social media, describing it as a carefree approach to coding with AI assistance.

The concept quickly gained traction, with Merriam-Webster Dictionary adding "vibe coding" to its database within weeks. Garry Tan, CEO of Y Combinator, reported that in the accelerator's Winter 2025 startup batch, 25% of the companies had 95% of their code generated by AI. This shift highlights the growing acceptance of AI in software development, where natural language instructions are used to create functional applications and games.

Vibe coding flips traditional software development on its head. Instead of meticulously writing code, developers can simply describe what they want to build, and the AI handles the implementation. This approach is exemplified by tech educator Matthew Berman, who noted that vibe coders do not write or read the code; they simply use it and refine it through conversation with AI assistants. Karpathy refers to this as the "Accept All" mentality, where developers trust the AI to generate working code without scrutinizing every line.

Beyond the technical aspects, vibe coding has spawned a distinct culture and aesthetic. The term "vibe" captures the emphasis on creativity, flow, and playfulness that practitioners bring to coding sessions. On platforms like TikTok and YouTube, videos of vibe coding often feature developers in moody lighting with RGB keyboards, ambient music playing softly in the background as they chat with their AI assistant. This fusion of coding with a chill, creative atmosphere has made programming appear more approachable and emotionally engaging.

Several high-profile tech figures have championed vibe coding, propelling it from a niche concept to mainstream discussion. Karpathy's influence, in particular, has lent significant credibility to the approach. Andrew Chen, general partner at a venture capital firm, described vibe coding as "brilliant and enormously frustrating," predicting that most code will be written by students and kids rather than professional software engineers. Amjad Masad, CEO of Replit, reported that 75% of Replit customers never write a single line of code thanks to AI features, underscoring how vibe coding principles are already changing who can create software.

However, not everyone is convinced of vibe coding's benefits. Critics argue that the approach could lead to poor code quality, security vulnerabilities, and skill atrophy among developers. Jonathan Blow, developer of the acclaimed indie game Braid, weighed in on the trend, stating that while getting stuff on the screen is easy, making the game good is hard. Security vulnerabilities represent another significant risk, as AI assistants do not inherently understand security best practices specific to a particular application. Blindly accepting AI-generated code could introduce security holes that might remain undetected until exploited.

There is also concern about skill atrophy. If developers increasingly rely on AI for even moderate coding tasks, their foundational skills might deteriorate over time. Debugging can become particularly challenging when the original code wasn't human-written. As Andrew Chen noted, while getting the first 75% of a project done with AI might be trivial, refining and fixing it afterward can become extremely difficult. Overall, the best approach thus far appears to be AI-assisted coding, where developers use AI to help them code but still debug errors and understand what the AI is doing.

In summary, vibe coding represents a significant shift in software development, where AI is used to generate code based on natural language instructions. While the approach has gained popularity and has been championed by high-profile tech figures, it also faces criticism regarding code quality, security vulnerabilities, and skill atrophy. As the trend continues to evolve, it remains to be seen how it will impact the future of software development and who will be creating the next generation of apps and games.