Wepresent our approach to agent modelingfor communication, and compareit to an approach to agent modelingfor other types of action. The comparisonshould be instructive since both approaches are implementedin the sameproblemsolving architecture, face similar application domainrequirements, and address the samegeneral problem of comprehension. Also, we show how for discourseprocessing,it is possibleto havethe benefits of viewingagents in terms of their befiefs and intentions withoutsacrificing real-time performanceand reactivity.