In support of the Soar/IFOR project's goal of providing intelligent forces for distributed interactive simulation environments, the NL-Soar project works toward the implementation of a full natural language capability for Air-IFOR agents. In this paper we discuss the design of that language capability (NL-Soar) and its integration into TacAir-Soar agents. In particular, we demonstrate how NL-Soar's linear complexity, interruptibility, and atomaticity of language processing provide language comprehension and generation processes that do not compromise agent reactivity.