The hype in agent research will not last forever. If the agent research community does not succeed in bringing its results into real-world application within the next two or three years it will share the fate of some other AI sections which are today thought to be just academic. Multi-agent systems applied to manufacturing are candidates for bringing the first breakthrough. To achieve that, methodologies l'or the design of agent systems for manufacturing by software developers are needed. This paper reports on the design process chosen in the MASCADA project, whose goal is not only to develop an algorithmic solution to manufacturing problems but also to show a transfer strategy for these results.