The notions of activity and context serve as unifying threads in this dissertation. As socio-cultural phenomena, both activity and context refer to complex and dynamic aspects of human existence. Prior context-aware computing (CAC) applications have failed to address this complexity and dynamism. We overcome these failures by defining activity and context relationally -- as co-dependent and mutually producing phenomena. Our definitions explicitly highlight the evolving impact that activity has on context and that context has on activity. We then translate our definitions into a novel computational model. This model differs from prior work in its ability to automatically detect and evolve representations of activities and contexts. Finally, we demonstrate the utility of our computational model by evaluating a prototype CAC application. This application is driven by, and enables interaction with, automatically generated representations of the user's activities and contexts.