The ability to automate the reinforcement of "working code" within LLM training means software is a mostly unique case where "good enough correctness" can be done at scale. LLMs are text generation machines, code is text, cli is text: only commands on computers are the kinds of text whose correctness is automated.

Its unclear what other jobs have a billion experts to-hand to grade LLM outputs. Its also unclear what other jobs would even benefit from such a thing.

For most others, it seems LLMs are basically tutors and mentors which existed anyway.