Ambient Intelligence (AmI) is characterized by dynamic ensembles of devices that offer individual services to the user in an unobtrusive manner. For more advanced services often the cooperation of devices is required which can be accomplished by automatic composition of services. AI planning is one possibility to realize service composition. We introduce an architecture containing a service composer that uses AI planning to compose services in AmI environments. The composer can be seen as an enhanced meta planner that uses service technologies. It makes use of three strategies for selecting planners and a heuristic to cancel planners. This design is based on extensive experimental evaluation in which we analyzed the behavior of planners. The overall goal of the design is to balance between runtime of planning processes, used resources, and erroneously not returned planning results.