Imagine five applications running on a computer, and each has a small AI model trying to optimize how that application uses the system.
Capable hardware can easily carry that extra weight.
But those models are still spending the user's computation.
That is what makes me question the architecture. If every application starts carrying a model to observe its resource usage, predict what it will need and make its own optimization decisions, we haven't removed resource management from the operating system. The OS is still there doing the actual scheduling. We have just added several smaller decision-making systems around it.
The OS Already Knows a Lot
