Apple Intelligence is a continuation of a mode of app automation I think of as "fracking".
For the OS to be able to understand what an app is doing, and to be able to poke deep into it to cause specific things to happen, the app has to make itself legible to the system.
Instead of a big monolith, the app has to create seams and sub-components, a process that is expensive and messy and I think of as "fracking".
The process can also be somewhat extractive, as the system can now squeeze apps more tightly and use them as a commoditized component in a larger user flow the system controls.
Apple has been investing in this approach for more than a decade, all the way back to public NSUserActivity's.
Each release of iOS they add more functionality for apps to frack themselves, more places that fracked apps can show up in the system.
Apple is not the only one doing this, by the way.
Google has been trying, although significantly less successfully–apps are far more wary about making themselves legible to Google.
This is not the only way to do this.
These approaches to app automation are with the app's assistance, but it's also possible to do an "over-the-top" style automation without the app being aware of what's happening.
The "over the top" model is one that Arc Browser, Adept, Multi-ON, Anon, etc are all attempting.
The over-the-top model is the only viable approach to integrating with existing experiences for entities that don't control the operating system.
The fracking approach starts with a high-quality set of apps that opt themselves in (depth over breadth).
The "over-the-top" approach starts with a low-quality but broad set of apps (breadth over depth).
Either approach might end up working!
Apple Intelligence is another incremental but significant step in the same direction they've established for a decade, now using the power of LLMs behind the scenes to increase their search quality.
This model can create user value.
But it has a low ceiling, because the app model presumes a privacy and distribution model that makes it:
1) challenging to launch an app speculatively in these kinds of assistive flows
Imagine asking Siri for help researching a trip to Hawaii. The next day you get an email from Delta trying to get you to buy tickets to Honolulu. As a user, that's a "wait, what?" kind of moment–it's not until after the fact you realize that behind the scenes the app was running and could have done who knows what.
2) impossible to run apps that the user hasn't already installed.
As a result, the various methods for integrating with apps of today can only ever achieve accelerated copy paste.
An outcome that is potentially good enough, but never transformative.
An alternate approach is to precipitate a new kind of software with different laws of physics that are more naturally amenable to speculative, combinatorial assistive user flows.