Andrej Karpathy's Vibe Coding U-Turn: AI Proponent Ditches Tools for Manual Coding

Even the Inventor of 'Vibe Coding' Says Vibe Coding Can't Cut It In a surprising admission, Andrej Karpathy, co-founder of OpenAI and proponent of "vibe coding," revealed that he wrote an entire open-source model by hand, contradicting his previous stance on relying on AI tools for coding projects. The model, called nanochat, is designed to let anyone build a large language model with a ChatGPT-style chatbot interface in hours, and Karpathy's decision to abandon "vibe coding" raises questions about the effectiveness of this approach. According to Karpathy, he wrote approximately 8,000 lines of code for nanochat by hand, using tab autocomplete. He attempted to use AI tools, including Claude Codex agents, but found them unhelpful and ineffective. This is a stark contrast to his previous views on "vibe coding," which he described as suitable for throwaway weekend projects. "It's basically entirely hand-written (with tab autocomplete)," Karpathy wrote in the project's documentation. "I tried to use Claude Codex agents a few times but they just didn't work well enough at all and were net unhelpful." Karpathy's decision to abandon "vibe coding" highlights the limitations of relying on AI tools for complex tasks. Despite its popularity, "vibe coding" has been criticized for oversimplifying the development process and neglecting the nuances of human expertise. Background and context: The concept of "vibe coding" emerged over a year ago, when Karpathy left OpenAI and began advocating for a more relaxed approach to coding. He argued that AI tools could handle many tasks, freeing developers to focus on high-level thinking and creativity. However, his recent admission suggests that this approach may not be as effective as previously thought. Additional perspectives: Industry experts are divided on the implications of Karpathy's decision. Some see it as a validation of traditional coding practices, while others view it as an opportunity for AI tools to improve their capabilities. "I think this is a wake-up call for the AI community," said Dr. Rachel Kim, a leading researcher in natural language processing. "If even someone who has been at the forefront of 'vibe coding' is now abandoning it, that suggests there's still much work to be done." Current status and next developments: Karpathy's decision to write nanochat by hand raises questions about the future of "vibe coding." While some may view this as a setback for AI tools, others see it as an opportunity for developers to revisit traditional coding practices. As the field continues to evolve, one thing is clear: the relationship between humans and AI will remain complex and multifaceted. In an email, Karpathy declined further comment on his decision, stating that he was focused on refining nanochat's capabilities. However, his admission has sparked a wider conversation about the role of AI in software development and the limitations of "vibe coding." *Reporting by Gizmodo.*