In this talk I will discuss the hazards involved in the design of large multi-agent systems (called Universal Information Ecosystems). These systems are a relatively recent phenomena and so there are relatively few guidelines for the potential designer. I will outline a number of interesting case studies where such systems have failed, and show that the design of these systems is actually not all that different from other design domains. I will then sketch a possible design methodology for dealing with these hazards.