The CASCADE Project was initiated to provide an environment in which faculty and graduate students could conduct research within an operational system. The idea was to build a small system that could easily be modified and extended to test new capabilities. Four research topics constituted the initial thrust of the effort: that were being

• How could SGML, hypertext, and DBMS capabilities be optimally merged to form a new kind of highly structured flexible net-based document construct?
• What kinds of protocols and communications techniques would best support the notion of network based documents?
• What kinds of interfaces would aid in the creation and management of these documents?
• What kinds of agents would aid this process and what kind of agent architecture would be most productive?

The project has made progress on all these fronts, and some of that work is in preliminary form now ready for publication. Other work is still being discussed. Two of the more interesting overarching themes have to do with the principles of interface design and the principles for agent design.

Related to the interface, we are now fairly comfortable with an architectural design overview that suggests that any task repertoire should be examined with four concepts in mind:

• Augmentation: In the classic sense with which it was initially proposed by Doug Englebart, augmentation refers to a design goal of an optimal split of the cognitive task between the human and the computational system.
• Information: This concept comes in its richest form from Shoshanna Zuboff's book "In the Age of the Smart Machine" and refers to the result of applying computation technology to some work process. In this context, information may be viewed as a term parallel to automation which is the application of power technology to the work process. Thus, the design goal is one parallel to the goal of automating some process -- it is to informate it. Without the rhetoric, one might view the simplest example of this principle as "Never ask a user to enter data twice".
• Visualization: The idea of visualizing things is easy. Essentially, anything that naturally exists in a three-dimensional space is pretty easy to visualize. Ideas are more difficult to visualize algorithmically. Much of the effort to date in CASCADE has played with visualization of documents and document spaces.
• Substitution: Substitution has to do with the use of agents. At one level, it is simply an extension of the concept of augmentation. The line is drawn to highlight the fact that augmentation is normally synchronous and non-autonomous. Agents are autonomous and may be asynchonous.

When I have more time, I will complete this next section on the agent architecture.