A Wireless Sensor Network (WSN) is a versatile sensing system whose hardware resources are scarce and have to be carefully used and hence adapted to face with their context and resources changing. Moreover, WSN could be used in unreachable or dangerous places, hence their adaptation should be remotely managed. In this paper, we propose a model-driven approach able to generate agents governing the sensor node in a WSN. The generation is automatic and the obtained agents are sent to the sensor nodes and executed over a mobile agents based middleware (i.e. a virtual machine). In this way, the sensor node logics can be automatically adapted without experiencing (from the user or the remaining system) service interruption. The approach will be developed in the VISION infrastructure to bridge the QoS management and the sensor nodes to guarantee best quality of service in case of context and resource changes.