The paper studies two closely related problems: how to support efficient message passing in large-scale agent systems given that agents are mobile, and how to facilitate the discovery of service agents in an open environment where agents may enter and leave. The Actor Architecture has been designed to study simulations of large-scale agent systems where agents obey the operational semantics of actors. We describe the solutions to these two problems that have been adopted in the Actor Architecture. The problem of efficient message-passing is partially addressed by using dynamic names for agents. Specifically, a part of the name of a mobile agent changes continuously as a function of the agent platform that it is currently hosted by. This enables the agent platform of a sender to use location information about the receiver agent in order to optimize message delivery. The problem of agent discovery is addressed by using a broker agent. Moreover, the sender agent may reduce the communication that is required between the sender itself and a broker agent by sending the broker an agent to localize the search for the appropriate service agents. In order to mitigate security problems, this search agent is very restricted in what operations it is allowed to perform and is transmitted in the form of a passive object. A description of the Actor Architecture is given, focusing on these two ideas and their preliminary evaluation.