Over a three-year period we designed and redesigned interfaces to a conceptually simple multimedia annotation tool called MRAS, the Microsoft Research Annotation System. We expected MRAS to be widely useful in education settings with little or no modification. In a series of deployments we encountered a surprising range of specific requirements. These included context- and content-specific needs. We ultimately had to shift from an application focus to a platform focus, from a single general-purpose tool to a toolkit to support asynchronous group interaction, with which task-specific applications could be built. Software that can be widely used with little or no modification has advantages for producers and consumers, but our experience suggests that as computer users become more aware of the flexibility of software, general purpose shrinkwrap software may fade away. Requests for MRAS from educators who would likely encounter similar problems suggests that these difficulties are not predictable. We discuss implications for designers, developers, and users.