Alternative hypothesis: people do gradually use these features, but HN became a weird anti-AI bubble for some reason.
I mean, many of the features had previous pre-AI attempts. They all sucked much more. This is the first time they have a remote chance of working (but in many cases, timidity or cheapening out is still making them barely usable - which is why 99% of the cases it's still much, much better to throw one of the three SOTA LLMs at the app to operate it (or on its data) from the outside).
Shitty LLM chatbots providing unhelpful (read: lies) customer service is even worse than some Phillippeens or Indian support with terrible scripting and half-broken engrish. At least you could deal with those fucks with demanding a manager or similar.
The LLMs will always do the worse option to customers, to extract more money. Burn'em all to the ground.