Our experience in the IDAS natural language generation project has shown us that IDAS'S KLONE-like classifier, originally built solely to hold a domain knowledge base, could also be used to perform many of the computations required by a natural-language generation system; in fact it seems possible to use the classifier to encode and execute arbitrary programs. We discuss IDAS'S classification system and how it differs from other such systems (perhaps most notably in the presence of template' constructs that enable recursion to be encoded); give examples of program fragments encoded in the classification system; and compare the classification approach to other AI programming paradigms (e.g., logic programming).