This is probably the most pragmatic take here and the overall responses support it. There's a handful of practitioners who are seeing some implementation success. They probably had significant domain expertise and are several iterations deep. Then there's a bunch of people who are experiencing the "AI iS mAgIc" approach where LLMs are being shoehorned into workflows with very little effort given to make the process actually… work (well)… to satisfy investor-driven mania. This growing-pains stage will pass eventually and we'll hit some equilibrium.

And even if the AI worked well enough (a claim to be proven by its proponents), companies wouldn’t give it access to be actually helpful. It’s all to save costs on routine calls, not enhance problem solving, the same way they’ve outsourced CS operators to body shops with a strict script.