Abstract

AI-native organizations increasingly create, transform, and act through representations generated by humans, AI systems, automated workflows, and external platforms. This condition creates a distinct architectural problem: how can organizations preserve meaningful correspondence among evolving representations of intent, decisions, work, evidence, outcomes, and learning when those representations are continuously reinterpreted and recombined? Existing literatures on enterprise architecture, organizational design, sociotechnical systems, information processing, knowledge management, organizational memory, cybernetics, digital twins, knowledge graphs, and AI governance explain important parts of this problem, but they do not provide a representation-centric theory of correspondence preservation. This paper develops Correspondence Architecture as an organizational architectural theory for preserving organizational cognition through governed representation networks. It argues that organizations sustain Cognitive Integrity when they preserve correspondence through canonical organizational objects, representation networks, organizational memory, traceability, evidence-based governance, assessment, learning, and bounded AI participation. The theory contributes to organization theory by shifting the architectural unit of analysis from structures, processes, or systems alone to the representation network through which organizational meaning remains intelligible over time. It contributes to AI-native organization research by explaining when AI participation strengthens organizational cognition and when it accelerates representational drift. The paper offers propositions, boundary conditions, and research directions for studying correspondence preservation in organizations where humans and AI jointly participate in organizational cognition.