> I'm kinda leaning towards the analogy that LLMs are to programming as textile machines were to the loom.

The difference is that textile machines reliably produce working cloth. LLMs do not, and indeed probably cannot, reliably produce working software. Instead they generate something which randomly has bugs sprinkled throughout such that a human needs to review the whole thing (which negates the gains). This would be like if a textile machine produced cloth that randomly would rip if you tried to use it for anything. Nobody would have accepted such a textile machine, just like nobody should accept LLMs.