Say I have a class called Foo and I realise, as I'm using it, that I want to add a method called doSomethingElse, I can just type the new method call and have appCode create the method for me. Ie:
Foo *foo = [[Foo alloc] init];
[foo doSomething];
[foo doSomethingElse]
doSomethingElse would be marked in red. I just have to hit command-enter, select 'Create method' and appCode creates the method signature in Foo.h, creates the method in Foo.m and puts the cursor in the body of the method ready to implement it.
This improves coding flow remarkably.
If doSomethingElse is a method defined in a category I expect to be loaded at runtime, or something I expect to be caught by foo's forwardInvocation method, having it turn red and bug me to provide an implementation in Foo is exactly the wrong thing to do.
It's the kind of little feature that really only makes sense in the context of a static language like C#, where a method call without a corresponding method implementation can be known definitively to be an error.
They've had to do similar things already with Javascript, Ruby, etc. Even Java has similar issues - for example you can use reflection to call methods. So it may mark a method as not being used in your project. But for 99% of cases it gets it right.
Some really good screencasts are needed to show what's possible with appCode - fortunately JetBrains does excellent screencasts for their other products and I'm sure they'll do these as it gets closer to release.
As for me, I'm a long-time IntelliJ user. Xcode 3 has been incredibly painful to use - even after taking a lot of time and effort to learn the keyboard shortcuts.
Xcode 4 is a major step in the right direction and started to actually make iOS coding reasonably fun. But it still can't hold a candle to the features available in this early release of appCode - it's awesome.