Although I'm enthusiastic about extensions, and many people share my enthusiasm, I've consciously (and with a hard-to-gather dose of restraint) decided to blog about the paradigms around packages and the package split first. The reason is to frame extensions properly. Extensions are a way to add features and functionality to the existing application, yes, but they are more than just a new technology feature in AX. They are an ENABLER to make the new paradigms and the strategy around packages successful. Developers should strive to use extensions to drive a move towards APIs and packages, not just to avoid over-layering (although the two are closely related). The goal is to move into independent loosely-coupled packages, and to attain that goal over-layering should be avoided, yes.
So what are extensions? If you think about customizations and Partner and ISV solutions, for the most part they add features and functionality to the existing application. I.e. they extend it. For this to succeed, they must be able to add fields to existing tables, execute code when certain things in the standard application happen, etc. Traditionally, this is done by over-layering the standard objects and code, and adding any needed new features. With extensions, the goal is to allow adding these additional features without over-layering, and even extending them from different packages. This seems fairly straight-forward from a metadata perspective. Indeed, adding a field to a table was already very granular by layer and model in AX 2012. And adding a form to a control was possible at runtime using personalizations. From a code perspective, however, things seem much harder. In AX 2012, eventing was introduced. One could add a pre- or post-handler to a method without technically over-layering the method. However, if for example a table didn't define a modifiedfield or update method one had to over-layer the table first to add the method, then add a handler to it. This obviously defeats the purpose entirely. And even when one added a handler or even an entirely new method to an existing class, you were technically still over-layering the class itself in a way.
So yes, beyond the expected tooling around adding fields to tables, controls to forms, etc., there are concepts to facilitate code extensions as well. First and foremost, events are everywhere. Although the traditional way of overriding a base method on a table or form or form control still is possible, most of those methods which were semi-events, are now ACTUAL events. A table for example has an OnUpdating and OnUpdated event, which allows you to subscribe to a pre- and post-update event without creating any methods on the table itself. A form button control actually has events for a click event, etc. These system-defined events are already numerous, and will be expanded further in future releases as the needs become clear.
So how about classes? Well, here there is already a difference between the original released version from February and the "May Update" (also known as "platform update 1"). Both versions have the concept of extension methods, which pretty much mimic the C# behavior of extension methods. With the May Update a new feature was introduced that allows a more broader concept of extending where the extending class can actually define new instance member variables it can use. Note that in both cases the extension cannot access private or protected member variables - it adheres to scoping rules, and makes sense when thought of in the API-sense.
Let's leave this wall of text for what it is, and start looking at actual examples of extensions in the upcoming blog posts.