If you want to consider the scope of agpl you first step is to consider the potential boundaries of a single work that are used over a computer network.
Take HN as a website. Should every resource be seen as a seperate work or should a whole page be seen as one work. Is every sub page a separate creative work, or part of a work called HN. Is the search button, which is actually on its own seperate domain name part of the HN website? From there we can move forward like talking about middle layer and db layers and ask if those are part of HN or simply aspects which HN communicates with. Is the loadbalancer, database server, web cache, and so on part of the creative work called HN?
Personally I would draw the line to be similar to traditional web development with presentation, logic and db abstraction defining the work in terms of a website, and everything around that being simply other works which the website communicates with. That however depend on the website being a classic website, rather than say an webified version of an application. A game, an office suit, a email client, all those might have a very different scope and what defines a game as a single work may be quite different from a website that display news items with comment threads.
> if Microsoft developers use network disks to access my software are they screwed?
That condition says that the terms of AGPL goes into effect if a use access a "derivative work" over the network. Microsoft using network disks internally does not matter, as that scenario does not have a defined user. If a user download a program over NFS then technically that is over a network but both AGPL and GPL would consider that "copying" in traditional sense.