The Internet of Things (IoT) encompasses a variety of technologies that enable the formation of adaptive and flexible networks from heterogeneous devices. Along with the rising number of applications, the amount of devices within IoT ecosystems is constantly increasing. In order to cope with this inherent complexity and to enable efficient administration and orchestration of devices, IoT platforms have emerged in recent years. While many IoT platforms empower users to define application logic for use cases and execute it within an ecosystem, they typically rely on static device references, leading to huge manual maintenance efforts and low robustness. In this paper, we present an approach that allows IoT platforms to autonomously and reliably execute pre-defined use cases by automatically discovering and selecting the most suitable devices. It establishes loose coupling and hence does not impose major technical constraints on the ecosystems in which it is operated.