Great, concise article. Nothing important to add, except that AI snake-oil salesmen will continue spreading their exaggerations far and wide, at least we who are truly in this business agree on the facts.
Books by programming theorists often When they define 'complexity' as 'anything related to the structure of a software system that makes it hard to understand and modify the system,' they miss a crucial distinction: the complexity of a supermarket is not the same as that of a telecom company. The primary factor in complexity is functionality and requirements to implement, followed by non-functional requirements, the restrictions of the IT environment, and then the structure of the software itself.
At this point, time becomes a crucial factor. You may end up with a creature that, after passing user testing and certification, has transformed into an unrecognizable monster despite your initial best intentions regarding length and clarity.
"Thanks for the article. I found it interesting, though the 'vibe coding' method isn't something I can apply in my own IT environment, mainly due to its complexity and technology stack.
The piece makes a strong case for why LLMs are so good at coding. It's clear they excel with the technologies and patterns most common in their training data—the same ones their own developers likely use. While their performance drops off with less familiar tech, it's undeniable that for certain types of tasks, they are very effective."
I think that beyond the language used, the article does have some points I agree with. In general, LLMs code better in languages that are more easily available online, where they can be trained on a larger amount of source code. Python is not the same as PL/I (I don't know if you've tried it, but with the latter, they don't know the most basic conventions used in its development).
When it is mentioned that LLMs "have terrible code organization skills", I think they are referring mainly to the size of the context. It is not the same to develop a module with hundreds of LoCs, one with thousands or one with tens of thousands of LoCs.
I am not very much in favor of skill degradation; I am not aware of a study that validates it in this regard. On the other hand, it is true that agents are constantly evolving, and I don't see any difficulties that cannot be overcome with the current evolutionary race, given that, in the end, coding is one of the most accessible functions for artificial intelligence.