Just like every other AI application they've come up with for end users, I cannot fathom how the AI aspect would be preferable to purpose-made code that does one thing without having to "think" about it.
How does the introduction of an "agent" who "might make mistakes" improve my user experience with... Literally anything?