Low-power sensors are increasingly becoming available, equipped with more energy-efficient processing and networking capabilities. Still, in order to accommodate the independent deployment and intermittent availability of such constrained devices, engineers often manually reconfigure system behavior for integrating sensors and actuators into complex and context-aware systems. The Multi-Agent Systems paradigm enables engineering systems where components can be deployed more independently and operate towards achieving their design objectives. In this process, they act autonomously and interact with others to perform context-aware decision-making without human intervention at run time. In this paper, we present autonomous agents implemented as low-power nodes that perceive and act in a shared environment through sensors and actuators. The autonomous agents on these constrained devices locally reason and act on the environment, and wirelessly interact with each other to share knowledge and enable more context-aware system behavior. The capabilities of our low-power autonomous nodes are demonstrated in a light-control scenario with two Belief-Desire-Intention agents. Our experiments demonstrate that running autonomous and social agents in low-power platforms incurs little overhead, indicating their feasibility.