The objective of this five-year project, funded through an NSF CAREER grant, is to study how social interface agents can conduct very long-term interactions with users spanning months or years of daily use and the impacts these interactions can have on user education, behavior change and overall well-being. A secondary objective is the development of a networked software architecture and experimental methodology to support very long-term human-computer interaction studies, in which new experiments and agent capabilities can be dynamically integrated into a running system serving a persistent group of human subjects. To be effective in both maintaining long-term interaction and achieving positive task outcomes, the agents developed will need to be able to interact naturally with users, forming social-emotional relationships with them over time. The test domain for this work will be physical activity promotion among urban older adult users. This work is expected to lead to a better understanding of how people and computer agents can optimally live and work together over extended periods of time.