Effectively procedural generation of all the graphics, which could make for some tiny apps (see the demoscene for a great example), and it's certainly a great idea that should be used more, but there's still at times a need to get pixel-perfect images of the right resolution due to the horrible effects that resampling has on the edge-cases. Ultimately the display is still digital/pixel-based, not analogue/vector-based unlike how CRTs used to work; and many things we view on the displays, like photos, are still pixel-based.
It's certainly a trade-off, I wouldn't pretend otherwise, but the vast majority of assets I produce at least are vector, which are then translated to various sizes of bitmap image, which is wasteful, painful and ultimately futile as there are still issues whenever a new device comes out (like the iphone 6).
The pixels of the screen are now so small they have almost disappeared, so I think there's an argument we have reached the place where going mostly vector makes sense.
Clearly some assets will always be bitmaps (photos), but so many others are not well represented as bitmaps (icons, text etc) - content is fine as bitmaps but controls and chrome are usually better as vector.
Photos seem like good examples of pixel-based data but it's really not. They are really often rescaled by an arbitrary real number compared to the device pixel. Antialiasing hurts vector images much more often than photographs.
