Predictably, developers are divided. While some see vibe coding as an evolution of higher-level frameworks, others argue that it is just the latest hype wave. After all, reliability issues in AI-assisted coding are well-documented.
Code-generating AI often confuses solutions, dependencies, or library references. A study published in arXiv found that commercial AI models recommended non-existent software packages 5.2% of the time, while open source models did so 21.7% of the time, posing a significant risk to production systems.
Even more telling is how the ‘godfather’ of Vibe coding, former OpenAI executive Andrzej Karpathy, doesn’t trust the technology enough, instead coding his own projects by hand.
Then there are the practical headaches. AI-generated code can be bloated, inaccessible, and opaque, making it harder to debug than manually written code. Thus, hard things like system design, debugging, trade-off decisions remain stubbornly manual.
<a href