Do people want that? I mean I don’t think it can discern if I say 15 or 50. Why would I leave that to chance that the ai properly grokked my message when despite what I’m guessing decades of work in the speech to text field, it is still pretty unreliable? Doing the task myself is trivial enough and 100% reliable.

In an ideal world in which LLMs behave as advertised, the idea is that you'd be delegating to the agent what you might otherwise be doing yourself. And just like when delegating to an intern who called you collect from a payphone, you may need to spell out "one-five" when you say 15 to make sure they don't hear 50.