> One thing with programming is that LLMs are actually pretty bad at it, and use of them for that purpose is driven entirely by hype

You haven't tried the latest gen of frontier models, I take it? These things aren't just hype.