New Delhi: The debate surrounding “vibe coding” has taken to X as developers list both pros and cons of the growing trend for AI to generate code based on simple prompts.
The term was popularized in early 2025 by Andrej Karpathy, an OpenAI co-founder. A style of development where the developer writes a natural language description without any particular structure into an AI tool like Claude and a code is generated by the tool. The fans state that it is much quicker to build prototypes and small projects.
The supporters cite speed and accessibility. They contend that now anyone lacking in formal education can send in a working product in a matter of moments. One of the engineers mentioned in a summary of the conversation went so far as to say that vibe coders make better products than traditional coders.
Critics do not have the same view. They provide alerts for bugs that make it to production, sloppy naming of generated objects and security issues. In one of the discussions, a user stated that 45 percent of the samples generated by AI contained vulnerabilities mentioned in the OWASP Top 10, which is a list of the top 10 web security vulnerabilities. This has led some teams to implement code review and governance policies for AI-generated code.
The summary, created using X’s Grok assistant, rephrases the argument as one that moves away from ‘heroic coding’ towards ‘industrial coding’. It observes apps created like this often reveal their sources when they need to be maintained. X’s disclaimer states that there may be inaccuracies in the summary and that it should be verified.
Every post is individually oriented to the mood. The user who posted the joke, “zam” (not his real name), got about 904,000 views and 70,000 likes with a statement about someone who renamed the practice “cuck coding,” as the developer was just sitting in a chair watching. The post got 6000 re-shares.
Then another person replied that they would get a better answer if they ran a free open source model on their own machine.There was another idle reply from Joel Guerra (@joelatwar), answering another person, that people could run a free open source model on their own machine and get a better result. It gained a couple hundred likes in that post.
Both teams seem far from giving up. Currently the debate isn’t so much about whether or not AI can write code, but who’s accountable for it when it breaks.
