Why LLMs Are Annoying When Programming Is an End in Itself
Sh0ny
4 августа 20262 min read
In short
In niche communities, it’s not just working code that matters, but also an understanding of how it works and what the author has learned from it. Let’s explore why LLMs are viewed here not as a way to speed things up, but as a way to skip over the most valuable part of the exercise.
Within niche developer communities, LLMs are increasingly viewed with hostility. The reason isn’t necessarily fear of new technology: for chess engine development, OSDev, LangDev, TxtDev, EmuDev, RLDev, the demoscene, and code golf, the process of mastering a complex field is often more important than a result that simply runs.
This changes the criteria for usefulness. In conventional development, generated code can be a good result if it solves the problem and passes testing. In hobbyist communities, this isn’t enough: respect is earned through understanding the underlying principles, the elegance of the solution, curiosity, and the ability to explain why everything works the way it does.
Hence the conflict with the typical LLM use case: describe a task, get a ready-made implementation, and move on. If the goal is to learn how to write a game engine, an emulator, or an unusual compiler, this approach can strip the activity of the practice itself. The working artifact will remain, but the journey that started it all will disappear.
There’s also a social issue. The author notes that early interest in LLMs in some communities quickly soured due to a combination of superficial understanding on the part of users and a sharp backlash from those who view such assistance as cheating. In closed technical circles, trust is usually built slowly—through years of participation and contributions to the shared body of knowledge. Showing up with ready-made code does not automatically earn that trust.
That said, the text does not imply that “LLMs are always harmful.” Rather, the model is more useful as an expert’s amplifier than as a replacement for them: a person with a deep understanding of the field can use it as a tool, verifying and guiding the result. But even expertise alone does not protect against model errors.
The main trade-off here is simple: LLMs speed up the process of obtaining a result, but in some hobbies, the result isn’t the main product. If you’re programming to master the craft, delegating a ready-made solution may not save time—it may actually mean giving up the very reason for undertaking the project in the first place.