This paper focuses on the challenge of building technical agents that act flexibly in modern computing and information environments. Flexibility is approached in terms of both cognition (ranging from reactive to pro-active) and sociability (ranging from isolated to interactive). It is argued that existing agent architectures tend to inherently limit an agent's flexibility because they imply a discrete cognitive and social behavior space. A generic constraint-centered architectural framework is proposed that aims at enabling agents to act in a continuous behavior space. According to this framework flexibility is not considered to be a property that can be directly implemented, but as a property that emerges as a result of an agent's continuous attempt to handle local and global constraints. The long-term challenge rised by the thoughts and ideas in this paper is to integrate various existing constraint-handling approaches and to valididate the usefulness of the generic framewo...