Introduction

Organizations increasingly act through representations. Strategies, decisions, operating models, policies, code repositories, dashboards, customer records, metrics, conversations, risk registers, AI-generated summaries, and lessons learned all make organizational meaning available for action. These representations allow an organization to say what it intends, what it decided, what work it undertook, what evidence supports its claims, what outcomes it observed, and what it learned. The focal problem of this paper is that these representations do not remain coherent automatically. They can drift, multiply, contradict one another, lose provenance, or become detached from the intent and evidence that once made them meaningful.

AI-native organizing intensifies this problem (Bharadwaj et al., 2013; Faraj et al., 2018; Kellogg et al., 2020; Raisch and Krakowski, 2021; Vial, 2019). In this paper, AI-native organizing refers to an organizational condition in which AI systems materially create, transform, relate, retrieve, summarize, or interpret organizational representations as part of ordinary work. AI systems no longer merely automate bounded tasks. They summarize conversations, generate plans, propose decision options, classify evidence, infer relationships, retrieve knowledge, draft recommendations, and transform organizational meaning across contexts. These capabilities may improve speed and coordination, but they also create a new architectural risk: representations can multiply faster than organizations can govern their relationships. AI-generated summaries can appear authoritative without preserving provenance. Decisions can detach from intent. Work can be optimized without evidence that it still implements a decision. Outcomes can be claimed without support. Learning can be generated without being incorporated into future intent.

Existing organizational architecture traditions provide important resources for addressing this problem (Henderson and Venkatraman, 1993; Lankhorst, 2017; Ross et al., 2006; Zachman, 1987). Enterprise Architecture offers frameworks for business-IT alignment, capability modeling, information systems, and transformation governance. Organizational Design explains roles, authority, coordination, and structure. Systems theory and cybernetics emphasize interdependence, feedback, control, and adaptation. Sociotechnical systems theory reminds us that organizational performance depends on the joint configuration of social and technical systems. Organizational information processing theory, coordination theory, knowledge management, organizational memory, and organizational learning each illuminate part of how organizations interpret, retain, transform, and reuse information. Digital twins, digital threads, ontologies, and knowledge graphs provide representation technologies for modeling systems and relationships.

Yet these literatures do not make correspondence preservation among evolving organizational representations the central architectural problem. Enterprise Architecture can align systems and capabilities without explaining how meaningful relationships among intent, decision, work, evidence, outcome, and learning remain intelligible over time. Organizational Design can coordinate roles and authority without specifying a representation architecture for preserving organizational cognition. Knowledge management can store and retrieve knowledge without guaranteeing that stored knowledge remains connected to the decisions, evidence, assumptions, and outcomes that give it meaning. Digital twin and knowledge graph approaches can model relationships without determining which organizational correspondences should be preserved, assessed, or restored. AI governance can constrain model behavior without defining the organizational architecture in which AI-generated representations participate.

This paper develops Correspondence Architecture to address that gap. Correspondence Architecture is a representation-centric organizational architecture for preserving the meaningful relationships through which an organization knows what it intends, decides, does, observes, learns, and adapts. It treats organizations as systems of represented meaning and coordinated action (Hutchins, 1995; Malone and Crowston, 1994; Weick, 1995). Its primary architectural object is not an organizational chart, process map, application landscape, data model, or software platform. Its primary architectural object is the representation network: the evolving network of organizational representations and the semantic, causal, evidential, governance, temporal, and learning relationships among them. The unit of analysis is therefore the representation network within the organization, and the explanatory focus is how correspondence in that network is established, preserved, assessed, and restored.

The paper builds on two prior theoretical claims. First, Organizational Correspondence refers to preserved meaningful relationships among organizational representations over time. Second, Cognitive Integrity refers to the dynamic organizational condition associated with sustained Organizational Correspondence. This paper extends those claims architecturally. It asks what architecture makes correspondence preservation more likely in organizations where humans and AI systems jointly create, transform, interpret, and reuse representations.

The paper makes four contributions. First, it identifies a representation- centric architectural gap in existing organization and architecture literatures. Second, it defines Correspondence Architecture as a distinct theory of organizational architecture, complementary to but not reducible to Enterprise Architecture, organizational design, knowledge management, knowledge graphs, or AI governance. Third, it specifies the mechanisms through which organizations preserve correspondence: canonical identity, representation networks, traceability, organizational memory, evidence-based governance, assessment, learning, and bounded AI participation. Fourth, it develops theoretical propositions explaining how Correspondence Architecture supports Cognitive Integrity, governance quality, learning, resilience, and coordinated adaptation in AI-native organizations.

The argument proceeds as follows. I first synthesize relevant literature and show why existing theories provide necessary but incomplete foundations. I then present the architectural principles that motivate Correspondence Architecture. Next, I introduce the conceptual model, define the theoretical constructs, and explain the correspondence-preservation mechanism. I then develop theoretical propositions. I conclude with implications for theory, management, architecture, AI-native organizations, and future research.

FIG-0003 depicts Research Positioning. It represents Correspondence Architecture, Enterprise Architecture, Organizational Design, Organizational Correspondence, Cognitive Integrity, Organizational Cognition, AI-native Organizations. The intended relationships are: Adjacent fields influence Correspondence Architecture without containment or replacement. The figure should be read as Not a Venn diagram and not a novelty-by-isolation claim. — Figure 1. Research positioning of Correspondence Architecture relative to enterprise architecture, organizational design, organizational correspondence, cognitive integrity, organizational cognition, and AI-native organization research.