The rational-empirical process was first proposed by the pragmatic philosopher Peirce to describe the manner in which a rational agent’s hypotheses are empirically validated and functioning in a social context. Sowa (2002) has updated and broadened these ideas as part of a knowledge cycle made up interacting abductive processes and a “knowledge soup” of micro-theories that are loosely assembled in lattices. Abductive process select chunks of knowledge from the “soup”, evaluate their relevance to the situation/problem at hand, and form locally consistent theories to predict observations. In this view the human mind is not a highly organized knowledge base so much as cognitive content that is assembled as needed into pragmatically consistent, local theory. This allows for a natural growth of knowledge flowing inconsistent, loosely organized, and in perpetual flux. The concept is captured by the term knowledge soup: fluid, lumpy, with adherable chunks of theories and hypotheses that float in and out of awareness. This formulation, which is centered in agent’s intentions and empiricallearning may be useful for designing cognitive architecture supporting interactive and inherently scruffy tasks, such as found in medicine and investigatory activities including planetary exploration. The paper connects Sowa’s framework with Bratman’s formulation of intention as a mental state that constrains future actions. Intention-based models of clinical guidelines, such as investigated in the Asgaard project, are discussed as fruitful avenues of research on the interaction between abductive reasoning and the weighing of alternative courses of action.