XML is being used in all of various ways. Originally a sort of
replacement for SGML for folks who didnt want the perceived weight of
SGML systems.
Long has XML been a data structure , a way to cast all sorts of data
in a package which is suitable to programs but also (sort of) eyeball
readable too, but I would'nt go so far as to say 'friendly', as some
have. (Want some evidence of this, try reading through CYC knowledge
base sometime, with just notepad or some other braindead text editor.
Sure if you use an editor that can "read" xml tags, do "colouring",
autospacing, yadda yadda then maybe the stuff is "friendly" to the
eye. (eye of a martian? human?) [BTW did you ever wonder why most dogs
dont watch TV? and no cats do? do you think maybe they cant actually
see the picture? do they care at all about barks, woofs and meows
coming out of some plastic (or wooden) box?]

XML is used in packaging the rule systems expressions (RuleML for
example) of the semantic web, the next step beyond just the plain ol'
data web. Their xml is used as a packager. By including stuff like
schemas(!) one can do lots of programmatical things and removce a lot
of what used to be clerk eye-powered checking.

XML is now being used as a quasi-graph-structured packager of content.
The memex system with its microfiche and so on was actually two mints
in one, freshens data into packets (like fiche) and also tastiest mint
of all because there was semantics / knowledge KEPT in the system such
that search was possible other than reading the fiche by eyeball!

Microsoft loves to hammer its chest about its directories but they
contain a paucity of information, just a few computer related things
like file size, date and few other things.

KDE4 introduces use of semantic data , relationships mostly, to start
with, which provide important meta-data to the system for associating
files (which might be pictures, movies, music, ...) and behaviours
(directed actions). Instead of trying to remember which 4th level deep
directory you put that thing of interest in 3 months ago, you can ask
the system to find it for you by typing in some info which it can be
use to "associatively" locate the item. Example, (show me) the picture
of that babe Yvette Mimieux in The Time Machine. If you have a
collection of 500 pictures of everything under the sun scattered
around 3 machines how many hours of searching by eye through countless
directories (if not backups) will it take for you to find that
picture, that you put in your system 3 or was it 4 years ago?

XML. Semantic Web, graph structured processing, slowly the future is
coming to today.