The problem of creating autonomous agents in a dynamic and unpredictable environment has led many to propose alternatives to the classical approaches of AI. Unlike traditional agent archetectures, these so-called ``nouvelle'' architectures tend to focus much more on good-enough behavior in unpredictable and noisy environments than on the ability to solve sophisticated tasks. This research is an attempt to start closing the gap between traditional and nouvelle styles of architecture by extending the behavior network algorithm, which was designed specifically for noisy and unpredictable environments, to handle more complex and sophisticated problem spaces. Specifically, this paper examines the problems with using behavior nets in domains with large numbers of features, objects, or locations which require differentiation, and discusses extensions which have been implemented to alleviate the problems. The advantages and limitations of these extensions are discussed.