Andrej Karpathy, the former Director of AI at Tesla, recently defined a term called “vibe coding” that explains an innovative way of coding that relies on sophisticated AI tools, especially large language models (LLMs). In his X post, Karpathy described that vibe coding is all about completely giving in to the “vibes” of the AI, accepting the exponential power of current LLMs, and forgetting the nitty-gritty details of the code itself. It is a transition towards how individuals interact with code and AI, where the boundaries between human interaction and AI-based output become indistinguishable.
Karpathy concept of ‘vibe coding’ and the evolution of AI
Karpathy’s “vibe coding” is a recognition of how sophisticated AI systems have evolved. In describing on X (formerly Twitter), he added that LLMs, like the Cursor Composer with Sonnet, are advancing to a degree that nearly eliminates the use of traditional coding mechanisms. Describing his own experience, Karpathy explained how he converses with AI tools almost in a passive manner—merely talking to them and having the AI handle the rest. This method eliminates manually typing code as well as keeping track of all the minute information in the program.
Karpathy’s approach to vibe coding: Simple tasks, AI fixes, and effortless projects
Karpathy went on to explain his individual experience with vibe coding, discussing how he deals with the AI tools so easily that he rarely even notices the keyboard. For instance, he just requests the AI to perform simple tasks such as padding adjustment on a sidebar, tasks that would otherwise involve more in-depth examination of the code. He takes it a step further by approving all the suggestions provided by the AI, without even checking the differences or changes made, relying on the AI to fix the problem. If there are errors, Karpathy tends to copy and paste the error messages into the system without comment, allowing the AI to take care of the fix.
This approach, while unorthodox, is functional for him, particularly for small, throwaway projects. Karpathy explained that he could still create functional projects without being forced to deeply explore the technicalities. “I just see things, say things, run things, and copy-paste things, and it mostly works,” he described, highlighting the seamless flow provided by the AI.
Karpathy’s honest take on the limitations of vibe coding
Though appealing, Karpathy did admit that vibe coding is not perfect. Occasionally, the AI tools are not able to repair some bugs, and in those instances, he has no choice but to make random changes until the problem resolves itself. Although this method may not be suitable for long, complicated projects, it works well for fast, experimental, or recreational work. Karpathy has a chuckle about using this method, understanding its flaws but enjoying its effectiveness for particular tasks.
Netizens’ responses to ‘vibe coding’
Karpathy’s concept of vibe coding elicited a broad spectrum of responses from X (formerly Twitter) users. One user commended Karpathy for being willing to adopt AI so comprehensively, adding that some programmers boast about writing all their code by hand without realizing that AI-supported developers might replace them in the near future. Another user noted that the degree of AI support in coding can be considered a continuum, with classical coding at one end and vibe coding at the other. Even Karpathy himself conceded that vibe coding is not yet in its ultimate version, implying there is still scope for development.
Other users appeared to be entertained by the idea, with one saying that they had been doing something similar for more than a year. Another joked, “What could possibly go wrong?” reflecting the experimental and unpredictable nature of depending heavily on AI for coding work.
Also Read | Genshin Impact Codes | NYT Wordle | Fruit Battlegrounds codes