>Not at all. It's not about simply replacing a DLL or updating using a binary patch.
>As a simple example imagine a 'MEF enabled' ordering system:
>
>At startup you give MEF a list of places where it will find bits that may be used by the application. This could be a concrete list of DLLs, or maybe you specify all DLLs in a specific folder or URI.
>
>Assume you've defined an interface named IProcessOrder. When the user hits 'Submit' MEF can search the catalog of parts, find one that implements the interface and give it the job. Let's assume that it simply writes the order detail to a SQL database. So far no big deal.
>
>But now the user comes to you and asks that when an order is submitted the application should also generate a packing list and notify the warehouse. You don't touch the existing code - just create another assembly which also implements IProcessOrder and drop that into a searched folder for the user who needs that functionality. The next time the user hits 'Submit' MEF finds both bits that handle IProccessOrder and you run both.
>
>Another user wants an email sent to the buyer when the order is accepted. No problem, just add another IProcessOrder implementation.

So, the magic remains in the second paragraph where something is put in the application in prevision of that.