Since recent years, the industrial world is facing a significant change from components and systems that are designed and configured for one well-analyzed scenario using well-established technologies to a demand for highly dynamic systems that continuously introduce the latest technologies and are capable of handling ever-changing use case environments. Thus, there is a need for new approaches on how to design and setup such systems. One of these approaches is Collaborative IoT, which focuses on researching the necessary designs, infrastructure, and technology to build ecosystems of loosely coupled, autonomous systems that interact in an ad-hoc fashion and establish service compositions and new distributed applications on demand. This paper presents a software architecture design for Industrial IoT systems to approach these targets as well as a reference implementation showing the feasibility of the architectural design in a Collaborative IoT smart factory scenario. The developed AiB architecture enables heterogeneous IoT systems to form task execution ecosystems that dynamically address new incoming tasks by creating service compositions across participating devices, which were not defined during design time.