It's only useful if the backend is wired to act on the request. LLMs handle the query understanding part, but it is still up to the backend developer to implem

It's only useful if the backend is wired to act on the request. LLMs handle the query understanding part, but it is still up to the backend developer to implement the functionality to fulfill the various possible requests. And the opaque nature of the interface gives the user no clue as to what requests are supported. We all experienced this problem with smart speakers, with their ever-changing features, and resigned to using an increasingly limited set of features.