In the execution of distributed algorithms on a network of processes, the actions of the individual processes are scheduled by their local schedulers or demons. The schedulers communicate with their immediate neighbors using shared registers or message passing. This paper examines an alternative approach to the design of distributed algorithms, where mobile agents are allowed to traverse a network, extract state information, and make appropriate modification of the local states to steer the system towards a global goal. The primary emphasis of this paper is system stabilization. Both single-agent and multi-agent protocols are examined, and the advantages and disadvantages of agent-based stabilization are discussed.