In order to manage fast changing market and customer requirements, production systems need to become more and more flexible. Therefore, the World Robot Challenge fosters the development of new agile manufacturing systems and evaluates them under tough conditions. In this paper, we introduce our service-oriented robotic system with which we participated in the challenge. The distributed software architecture covers four components: Robot Controller, Vision Module, Configuration Server and Process Controller. In order to enable a fast hardware setup, we used one collaborative robot and developed additional jigs and tools to solve the tasks. During development and testing under competitive conditions, we experienced the immediate benefits and drawbacks of our distributed architecture and its individual components. As a result, we can improve the partitioning of our services. The object detection based on a Convolutional Neural Network had to deal with false classifications due to changing lighting conditions. The combination of rough position estimation by using a global camera and a precise calibrated wrist camera for the visual servoing has turned out to be an essential element.