Automatic Guided Vehicles (AGVs) are fully automated vehicles that are able to transport goods in an industrial environment. To cope with new and future system requirements such as flexibility and openness, we have applied a situated Multiagent System (MAS) to develop a decentralised control architecture for AGV transportation systems. In this paper, we give an overview of the software architecture of the system and we zoom in on two specific concerns: transport assignment and collision avoidance. We discuss the evaluation of the software architecture and the test results obtained from realistic simulations and a demonstrator system that we have developed. The architectural design and development of this real-world application teaches us that the primary use of a situated MAS comes from the way in which it structures the software. In particular, the set of adaptive agents that coordinate through the environment allows us to shape the software architecture of the transport application to provide the required functionalities of the system and achieve the important quality goals of flexibility and openness.