i think LLMs have a particular appeal for engineering types who like automating things and have successfully, in the past, automated lots of tasks that used to be manual. it gives the illusion that they can finally, FINALLY, automate those uniquely human tasks that have eluded them thus far. and it's the whack-a-mole process described in the post. it's always 85% there, just tweak one more thing, add one more agent, fine tune one more model.