Conventional object oriented middleware platforms rely on the notion of remote interfaces to describe distributed services. This notation is very similar to the one currently used in centralized systems, which increases the productivity of programming. This paper is founded in the observation that remote interfaces foster a programming model that ignores the differences between local and remote interactions. This can result in distributed applications with poor performance, that are not robust to failures, and that can not scale beyond local networks. Therefore, we propose that remote interfaces should be accompanied by the specification of tactics that deal with typical events in distributed computing, such as concurrency, partial failures and high latencies. The paper proposes a tactics definition language and describes the implementation of a middleware system that supports this language.