User interface management systems (UIMS) have established themselves in both research and commercial arenas. We present several generations in UIMS evolution and discuss some problems of the early generations. In particular, we discuss the problems of a gap between methods used by behavioral scientists and computer scientists during the process of building interfaces. We present an empirical approach to begin bridging this gap and results of our preliminary observations: a human-computer interface development life cycle and recording techniques for interface development, as well as UIMS needed to support them. We conclude with future directions for the evolution of UIMS.