Probably one of the most serious problems
with AS/AD analysis is the assumption of the 
independence of the two curves.  In reality we
would rarely see one of them shifting without 
the other one doing so also, especially if the
one shifting is AS.  But this independence has
also often been overstated at the micro level
as well, although I rarely see anybody losing
sleep over it.  But one of the underpinnings of
the identification problem that lies at our
inability to figure out whether or not bread in
19th century Britain was really a Giffen good or
not is precisely due to the problem that the
producers of the bread were also major consumers
of the bread.  I would also note at the micro
level, that in many industries changes in supply
take a lot of time, so the whole ceteris paribus
assumption for constructing such curves is just
ludicrous in reality.
    At the macro level, I have no problem with
engaging in the following kind of exercise.  Let
both M and V be fixed, hence a fixed nominal GDP
with a hyperbolic AD curve a la Fred Foldvary's
(although for teaching I prefer more behavioral ones).
Pick any old AS you want, vertical, upward-sloping,
with zones, whatever.  Let it suddenly shift backwards
due to a negative supply shock, an oil price shock
(was this not the case that brought AS/AD analysis
into the textbooks?).  Let the AD curve remain fixed,
no changes in M or V.  One then moves backwards up
along it to observe the classic stagflationary outcome,
no need to worry about disequilibrium effects or any
of the rest of this, much less where this is coming
from with regard to the (or an) ISLMic analysis.
      Artificial?  Yes, but not all that much more
so than what is done at the micro level, especially
given that the foundations for standard utility 
maximization that supposedly lie behind our micro
demand curves are completely in the trash can, with
only the textbook writers not dealing with this as
the theorists have yet to pick up the garbage and
figure out how we do those things in the light of
what is now known about how people behave (again,
see George Akerlof's lecture and related work in
behavioral econ).

Barkley Rosser