We contend that current approaches to supervisory control of complex systems are inherently insufficient. In this paper, we provide a collection of anecdotal evidence supporting this contention, and claim that combining innate human flaws with sophisticated but brittle automation leads to a slew of problems. We briefly outline future research directions that we believe will lead to more robust, capable, and efficient autonomous control systems that team effectively with humans. In short, while improved human interaction systems can help, the fundamental change required is the addition of another partner in control, a model-based intelligent system that can turn "three’s a crowd" into "four’s a team." The anecdotes and scenarios described below have all happened, many during our visits to petroleum re- fineries and chemical plants around the country.