I think fragments do this very well, its just that the *way* that certain aspects of the system (as it relates to UI) are implemented that shows lack of a unified vision.

For example.

Android has a tertiary language called Renderscript to produce very quick graphic code. Now, this is cool and all, but requires programmers seeking performance not only leverage Java to produce their works, but also C (C99 to be precise), compile to a renderscript file (LLVM bitcode), and then tie-in to their java application. Wha???

It seems clear that renderscript is the brainchild of a small group that sees its benefits, but that the Dalvik VM is the brainchild of another group and they are being duct-taped to one another. Now I'm not saying that interoperability is bad; it's quite good. I am saying that why on earth should you need to write your program in possibly three languages (Java, JNI Native, and Renderscript) to produce well running code? This is the type of complexity that can be avoided.

Renderscript is a very cool concept, but why, if you're compiling LLVM would you call it Renderscript, confine it to C, and limit its main functionality to UI graphics? Why not compile Java to LLVM bitcode, and work on perfecting a better LLVM compiler rather than Dalvik? It shouldn't be Renderscript, it should be Generalscript, and the language language should be arbitrary -- that's the point of LLVM!

My point is this: If Renderscript is so fast, and can be called and call Java in the Dalvik VM, offer it as an alternate to Java and not a specialist tool.

It makes little sense.

Additionally, fragments are cool, and views are cool, but put some sparkles on it for goat sake! There should be pre-baked APIs for doing cool animations and transitions and otherwise making the UI look pretty whilst being pieced together in XML. There should be layouts that are proven to work with different form factors easily selectable for developers so all they have to do is go to a page, see a layout that looks cool and would work with their site, watch a video of it in action, download it, and fill in the code behind the interface, and reap the benefits of cross-platform UI goodness.

Lets face it: tablet UIs are all pretty darn simlar, why not take the work out of it, and make it easier for developers? In my perfect world, developers would be able to select portions of the UI from examples, piece them together, and fill in the logic.

Currently, this is not how things are done. Its not terribly hard now, but it's different enough to make it a learning experience to jump in, which may turn some folk off, or increase the amount of time it takes to deploy an app.