Why hobby coders reject LLMs: the craft is the point, not the output

Prompted by a chess-engine GitHub thread, Fogus argues that niche hobby programming scenes — OSDev, LangDev, EmuDev, RLDev, code golf, the demoscene — increasingly reject LLM-generated code because their entire reason for existing is the difficulty of the work. In these communities the real product is mastery, not a program that happens to run. Standing is earned slowly through years of participation, elegant code, and shared domain knowledge, so tools that hand you a finished answer are seen as skipping the only part that matters.

He notes that even good-faith attempts to bring LLMs into these scenes soured fast, undermined both by practitioners who lacked deep understanding and by a vocal faction that treats LLM use as outright cheating. He concedes these groups have long been marked by heavy gatekeeping and slow progress, which makes the appeal of an instant shortcut — and the backlash against it — easy to understand.

Fogus’s own position is that an LLM works best as a force multiplier for someone who already understands a domain, a lever rather than a substitute. In a context where the whole exercise is learning, generating the finished artifact doesn’t make you a craftsman; it strips away the craft itself. He closes with a caveat: expertise offers no guaranteed protection against being misled by these tools.