Machine-to-Machine (M2M) systems are smart networks composed of multiple devices sensing or acting in the physical world. These devices interact together to provide data to value-added services or applications developed by various stakeholders. M2M infrastructures aim at being deployed in city wide area. They have to handle and support an increasing number of services and applications. A dynamic sharing of M2M infrastructure between applications is thus a strong requirement. To answer to such a challenge, we propose to use the multi-agent paradigm and related programming languages. Adopting a multi-agent oriented programming approach, we define an adaptive and agile layer to govern M2M infrastructures supporting different applications.