This is true up to a point. To my understanding, in a sandboxed environment a third party AI can only work if it is incapable of choosing which application should execute the intents. All it can do is determine the intent and pass it along to the OS, which then presents the user with a confirmation or, in the case of multiple apps receiving the intent, a tiebreaking interface.
A possible workaround would be a cumbersome two-way permissions system ("can app X access app Y?" [and, for purposes of apps asking the AI to ask follow-up questions] "can app Y access app X?", ad nauseam), but this is something of an impractical solution, because the AI would need to be granted permission every time the user installs a new app to access that app.
So yes, Apple makes it the job of the OS. This is a design compromise that mitigates the risk that the user is overburdened with confirmation dialogs, choice dialogs, and/or permission dialogs. And for an AI that just works, I would say this is pretty key.
Android has a centralised 'account' system, to access things like twitter, facebook, gmail etc. These can be used for applications to connect to these services. Android also provides permissions for accessing contacts, call logs, calendars and latitude. So in some sense, Google have had to make Android do this as well, but it would be hard to do it with the way iOS currently works.
A possible workaround would be a cumbersome two-way permissions system ("can app X access app Y?" [and, for purposes of apps asking the AI to ask follow-up questions] "can app Y access app X?", ad nauseam), but this is something of an impractical solution, because the AI would need to be granted permission every time the user installs a new app to access that app.
So yes, Apple makes it the job of the OS. This is a design compromise that mitigates the risk that the user is overburdened with confirmation dialogs, choice dialogs, and/or permission dialogs. And for an AI that just works, I would say this is pretty key.