The proliferation of conversational agents (CAs) promises efficiency and quality improvements while enabling a more seamless integration of technology into service encounters. However, it remains un- clear how CAs should be designed to provide the optimal experience for the key users: clients and frontline employees. Based on qualitative research with those key users, this study delivers a vision of an adaptable CA. It proposes a differentiated approach toward the design of CA: there is no "one-size- fits-all" design regarding the level of social presence, autonomy, or agency. The analysis reveals three tensions in user expectations leading to inconsistent design requirements for CAs. To resolve those ten- sions, CAs should be adapted to the changing context of a service encounter considering the appropriate level of autonomy, task complexity, interpersonal intimacy, and social role of the CA. The study con- tributes three design principles emphasizing the importance of the context for which a CA is designed.