I want to be able to drive my car myself and I don't want autonomous vehicles on the road unless they've been proven to be safe. I figure by 2030-2040 is enough testing.
Can we apply the same standards to human drivers too please?
Collaborative editing is often easier. This isn't necessarily a property intrinsic to online office systems, but offline ones are typically intended for offline editing. Even with a decent revision control system (is there one that can merge OO.o or MS Office docs? No idea), you periodically get sets of changes from other people and have to merge them. If you've got something that allows live editing by multiple people, you can see what other people are doing at the same time as you and avoid conflicts.
There's no reason that an office application couldn't support this, it just seems to be a feature they haven't implemented. You'd probably want a single server for your organisation that would track all of the changes (allowing every desktop to accept connections for peer to peer editing would give network security people nightmares, especially considering the security record of MS Office), and it would be great if the server could push change sets out to some revision control system so that they could be synchronised with other documents (maybe push live editing into a branch and then have a merge step as part of hitting save).
You can fix it, but I agree that it usually puts it in the worst possible place. The problem is that TeX uses an elegant dynamic programming model to determine where to break lines in a paragraph, but uses a greedy algorithm to do page layout. Why? Because the PDP-10 didn't have enough RAM for the dynamic programming tables that would be required to do elegant page layout on a typical document. On a modern computer, even if it takes 2-3MB for the tables, you most likely have a single image in the document that is bigger than that (in early TeX, images had to be added afterwards in a separate compositing phase after you sent the typeset document to the printer, because computers weren't powerful enough to handle nontrivial images).
I tried implementing the TeX linebreaking algorithm for page layout in some naive (unoptimised) Objective-C a few years ago and ran it on a 900-page book that I'd written. Even then, it took under a second to run on the laptop I had at the time. There's no reason not to do it now.
With your bare hands?!?