The first versions of this kind of thing are often "declarative"; you push your data (with strong limits) to it, and it handles the rest. This way a level of performance can be guaranteed.
It will be a few generations of hardware before the system is more "free."
I believe the author is a degree beyond wishing the apps were more powerful on limited hardware, c.f. references to "iPad-ization" and Swift playgrounds.
Succinctly, I believe they'd nod along to an example of "It. Blew. My. Mind. as a 6 year old to get bored, click around, and find image files that appeared in Civilization II."
https://en.m.wikipedia.org/wiki/Retained_mode
The app is running on the device, but the app is pushing the commands to render 3d models to the system.
One example is that right now Apple makes a distinction between Immersive (AR, digital objects superimposed in the real world) and Fully Immersive (VR, fully digital, real world is no longer visible).
In Fully Immersive mode, a developer gets full control over the entire screen and can draw using a custom rendering engine on top of Metal.
But in Immersive mode, RealityKit handles all rendering, and you’re restricted to whichever rendering features RealityKit exposes. There is some support for custom shaders, but it’s quite limited.
Unity solves this issue using PolySpatial, which translates Unity’s scene representation to RealityKit.
In that sense I can imagine (or hope) that Apple opens up the platform more. Similar to how it added the AppStore to the iPhone after one generation.
But it will, unless some drastic internal push and change in direction happens, remain iPadOS-like.