From several years we are developing robotic multi-agent systems according to well defined design methodologies. These methodologies evolved over time because of the changes in the operating environments (robotic hardware and software platforms) and specific missions accomplished by our robots. In the last four years we used PASSI (Process for Agent Societies Specification and Implementation) obtaining good results but, the growing experience and day by day accelerating changes in requirements suggested us to find a new and more versatile approach. In this context we developed the Agile PASSI methodology discussed in this paper; it is an agile process built up capitalizing all the experiences done with PASSI and its supporting tools some of which have been adapted and reused in the new process.