Although I haven't tried it, vibe coding is using an LLM to code. The erotic fantasy is that this now means you can get code done without actually knowing anything. Just give the right prompt and the AI spits out working code. In practice, this leads to a maximum technical debt approach to coding. For example, an AI might have several bits of nearly identical code, rather than making a proper reusable function.

The current batch of LLM AIs just fail for tasks that require precise thinking. For example, these LLMs just fail when they try to play chess, hallucinating illegal moves. Now imagine that same chatbot trying to write code.

https://www.youtube.com/watch?v=6_ZuO1fHefo&list=PLBRObSmbZluRddpWxbM_r-vOQjVegIQJC&index=1

I guess a lot of software engineers really are so bad that a buggy AI is actually an improvement.