I'm currently working on a "little a" agile team. Upper management doesn't really care what our process is, as long as they get their quarterly release with (most) of their priorities.
As soon as upper management catches wind, you'll need to justify every little process flow change to them, or they'll start demanding regular reports with burndown charts and plots of velocity over time and crap like that. At which point, you're done. Things officially suck forever.
That isn't to say that I think Scrum coaches and whatever have no value to offer. I've done a 1-week Agile training course, and I thought it presented some really great ideas. But their potential value is severely undercut by the fact that there's no way to bring one in without attracting the attention of the mothership.
I think that's a management problem you'll run into with any development methodology you use. There are a lot of management types that slavishly focus on metrics whether or not they make any sense, and they can do tremendous damage to any organization.
Give those people a tool and they'll find a way to misapply it, whether it's velocity, or performance reviews, or LOC, or whatever.
Prime example of this was stack ranking at Microsoft. How much productivity was lost due to the political maneuvering and backbiting caused by their misunderstanding of bell curves?
As for 'i can get something in 2 weeks rather than waiting for 2 years', that's about not waiting 2 years to find out that you've been building the wrong thing or that the software sucks. Maybe as you deliver features you'll find out that the application is good enough in 18 months. Maybe you'll find out after several months of sprints that the product owner is bad at getting requirements and you need to do something different. Or maybe you'll find out that it is taking longer than original estimated to deliver functionality so important dates need to be adjusted accordingly.
In the old waterfall model, all that bad news may not come out until 2 years later when the software is released. And in the old model the customer is not getting useful software until 2 years out, rather than much sooner with a minimally viable product.
Agile is not a panacea and any methodology will fail if attempted by dysfunctional/incompetent groups of people. But if there's one benefit of Agile, is that everyone should find out much sooner that things are going awry.