In Machine-to-Machine (M2M) systems, multiple devices (sensors, actuators), situated in the physical world, interact together to provide data to added value services. In the SensCity project, the proposed M2M infrastructure, to support cityscale application, must be able to support the increase in the number of services and applications that are deployed on it. It is thus necessary to share the infrastructure use dynamically between them. In this paper, we propose the use of multi-agent technologies to define an adaptive and agile layer to govern and adapt the M2M infrastructure to the different applications using it. We illustrate our proposal within a smart parking management application.