Computers knowing about focus is quite a compelling future direction. There are lots of really good possibilities. But also a few problems.
For instance, If you can track focus, you can make the monitor seem larger than it really is. Just scale everything that isn't being looked at down a bit. As the gaze shifts towards other objects, shift things slowly around and overlap the enlarged window over other background windows.
You can make focus-dependent shortcuts. Imagine vim with nouns and motions that can refer to and act on the focus point. `ytF`: yank-to-focus, where F is a motion from cursor to focus. Lots of rich possibilities there.
The only major problem I see is that you can't really share work easily (unless you open it up to multiple focus points somehow.) Also, it would be frustrating to have to look where you're typing. I'll often look at something else while typing just before switching tasks. I'm going to start paying attention to my focus more to see if there are other potential pitfalls with the computer knowing about it and changing modes in response.
In general I like the possibilities opened up by having focus. Heck even with a traditional mouse+keyboard, the extra data would help the computer understand us better. It might be more suited to a VR desktop, where the 'multiple-viewers' problem does not exist.