Many practitioners and researchers in workflow systems have stressed the importance of flexibility in such systems. On the one hand, organizational control requires detailed modeling of workflow processes and stringent procedural guidelines. On the other hand, dramatic organizational changes and frequent exceptions demand system flexibility. Few satisfactory solutions have been provided thus far to deal with this issue of flexibility because workflow systems developed using conventional software techniques are costly to modify and offer limited flexibility in general. However, recent progress in intelligent agents, especially autonomous and adaptive agents, seems to hold promise in providing increased flexibility in software systems. In this paper, we discuss the problem of flexibility in workflow systems, overview the features of intelligent agents, survey their applications in workflow system prototypes, and propose several research issues on the development of agent-based workflow systems to achieve increased flexibility.