2.1) Solving information problems. What you describe.
2.2) Solving physical/mathematical problems. Think signal processing. Think physical simulation. The bulk of the code is based on some mathematical theorem or linear algebra that is just too complex to be calculated by hand. The UI is just the surface.
The 2.2 group seems a bit more selective. Applications like audio workstations, simulation grids or wheather forecast centers need this. Universities use this for research.
While these two groups aren't completely exclusive, I'd say many CS graduates lack the engineering/mathematics/physics prowess needed for the problem spaces encountered in 2.2.
You claim that GMail isn't complex. Jeezus, a tiny % of programmers are capable of recreating that well. Or Microsoft Office! Go get the source code to OpenOffice and say that again. Think about the problem a bit harder and you'll realise it's a freaking compiler that's far more complex than most programming languages.
Far, far, far, far more complex than Dropbox.
I bet the hard bit of Dropox was the seamless UI integration with the OSes and not the routing of data. The data routing was probably solved in an afternoon. That was probably the fun bit.
You seem to have no clue about what is actually hard in programming. Hard problems, pffft.
It's as if I went to projecteuler.net and looked at some code and went 'oh, look, implementing algorithms is easy. Any A-level maths student could do that...'
Graphic designers generally aren't capable of programming or maintaining a large CRUD app, and when they try what usually happens is that some professional has to come and gradually rewrite the whole thing. Without breaking anything. That's hard.
If they are capable then they usually call themselves a programmer, and rightly so. And if they wanted to solve the 'hard' problems, they could. They'd just have to learn about a different area of computing. And I'm mightly jealous of them for having mastered two hard subjects instead of just one.
So next time you want to declare that 90% of programmers could be replaced by people with no programming skills, go look at some source code and then shut up.
Hearing someone say they're a software architect or that a company has a software architect is usually a big red flag to run like hell.
Also a large code base is identical to a complex problem. Check the definition of complexity. A complex problem can be broken down into simple components.
Finally the best glue is not complex, it's simple. Complex glue tends to result in brittle magic, a poor abstraction, like traditional asp.net compared to something far more elegant and smaller like ruby on rails.
1) Your average database/object driven application. There is user input. This translates pretty directly to output. There may be some interesting algorithms in between, but it's pretty much UI programming.
2) Solving difficult problems. A logistics routing application. The bulk of the code is based on graph theory, reducing the problem perhaps into smaller segments that can be solved in different ways. The UI is just the surface.
There is a need for both types of programmers. The first type are increasingly graphic designers who decided to learn PHP or some similar situation. There are tons of great applications out there that really aren't that complex (except when scaled to extreme proportions - ie Twitter). Wordpress, Gmail, even Microsoft Office isn't terribly complex (at least, the 90% of it that everyone uses).
The second group seems to be a bit more selective. Applications like Dropbox, which need to route data in the most effective way, or Adsense, in which profit needs to be optimized in choosing what ad to display as a page loads. These are complex problems, with complex solutions. That's where O(n) analysis is important. That's where an understanding of graph theory and established algorithms is critical.
While these two groups aren't completely exclusive, I'd say many self taught programmers find themselves in the first group.