The title of Stefan's talk is great: The Unreasonable Effectiveness of Multiple Dispatch. He gives a nice example of composability: how you can throw a new type into an existing algorithm and it just works.