Publisher Summary Each person, group, or program will have its own information space that is always quietly evolving behind the scenes and according to the owner's wishes. These information spaces will be simple to share, thus allowing us to easily trade data and build new information spaces out of old ones. Information spaces will even become predictive by learning owners' data habits, and will thus offer highly valuable data that was not even requested. The software layering and the promises go on. Agents, armed with user profiles and declarative user requests, will find just the right stuff, and then use mediators to extract, integrate, and reformat data. Agents will often animate data, making it sing and dance on the screens. This is all part of a widespread phenomenon that can be seen throughout computer science. Consider the database world and the problem of extracting data semantics. The spectrum of elegant-to-engineering solutions is decades old and is based on the creation of metadata—schemas, annotations, structured terminologies, etc., and ways of manipulating that metadata. The elegant solutions started coming out in the '70s and '80s. When these proved ineffective, far less aggressive goals were laid out, and less functional but far more useful solutions emerged and have found their way into the commercial world.