In short
It is not uncommon for a non-technical specialist to find it difficult to explain the difference between code that simply works and code that meets quality standards. This issue is particularly relevant when working with code generated by neural networks. By drawing analogies from the field of writing, we can clearly demonstrate why “works” does not always mean “is done properly.”
In the world of software development, misunderstandings often arise between developers and people who aren’t involved in programming. For “vibe coders”— as the author calls those who see only the superficial result — the main thing is that the program performs its function. But the reality is much more complex: “works” does not equal “optimally” or “well.”
Modern neural networks are capable of generating functional code. However, as the author notes, such code is often:
The challenge lies in how to convey this idea to someone who lacks in-depth knowledge of programming. How can we explain that code—even if it performs its tasks—can be significantly worse than code written by an experienced developer?
To illustrate this idea, we can draw parallels with written text. Anyone can assess the quality of written content, even if they aren’t a philologist or an editor. Imagine two texts, both of which convey the same information:
Formally speaking, both texts “work”—they successfully convey the information. But the quality of perception and understanding, as well as the further use of these texts (for example, for translation or adaptation), will be radically different. The same thing happens with code: “working” code can turn out to be a nightmare for maintenance and development.