This paper describes principles and practices of knowledge engineering that enable the development of holistic language-endowed intelligent agents that can function across domains and applications, as well as expand their ontological and lexical knowledge through lifelong learning. For illustration, we focus on dialog act modeling, a task that has been widely pursued in linguistics, cognitive modeling, and statistical natural language processing. We describe an integrative approach grounded in the OntoAgent knowledge-centric cognitive architecture and highlight the limitations of past approaches that isolate dialog from other agent functionalities.