Introduction

Organizations increasingly rely on AI systems that do more than store data, retrieve information, or automate bounded routines (Brynjolfsson and McAfee, 2014; Bharadwaj et al., 2013; Faraj et al., 2018; Iansiti and Lakhani, 2020; Russell and Norvig, 2021). AI agents can summarize organizational memory, draft options, coordinate work, classify evidence, monitor exceptions, propose actions, execute tasks, and participate in learning cycles. As these systems become embedded in work, organizations begin to distribute cognition and action across a mixed population of human actors, AI agents, collective actors, roles, representations, workflows, and governance arrangements.

This is not simply the familiar problem of adopting a new technology. Ordinary AI adoption asks whether a tool can improve an existing activity. AI-native organizing asks how an organization should be constituted when non-human computational actors participate in the preparation, execution, and monitoring of organizational action. The question is not whether AI systems can act in a technical sense. Many can produce outputs, invoke services, or complete delegated steps. The deeper question is whether such action is organizationally legitimate, accountable, observable, reversible, and adaptable.

Conventional organizational forms provide important starting points (Weber, 1947; March and Simon, 1958; Burns and Stalker, 1961; Lawrence and Lorsch, 1967; Mintzberg, 1979; Puranam, 2018). Hierarchy clarifies authority and escalation. Matrix and network forms address interdependence. Platforms coordinate modular participation. Operating models connect processes, governance, people, and technology. Enterprise architecture describes relationships among strategy, capabilities, information, applications, and technology (Ross et al., 2006; Zachman, 1987; Lankhorst, 2017; The Open Group, 2022). AI governance frameworks articulate principles for responsible design and use (Floridi and Cowls, 2019; Jobin et al., 2019; Shneiderman, 2022). Multi-agent systems model coordination among computational agents (Wooldridge, 2009). Yet each tradition leaves part of the AI-native organizational problem unresolved. The problem is not that these theories are obsolete. It is that they often treat the constitutive basis of organizational action as already settled. Hierarchy assumes that authority ultimately flows through human roles. Networks explain distributed coordination but often under-specify legitimate authority and revocation. Enterprise architecture links organizational and technical structures but does not by itself constitute actor authority. AI governance addresses model risk and responsible use but often treats organizational design as context rather than as the object of theory. Multi-agent coordination explains agent interaction but does not settle organizational legitimacy or accountability.

The central problem is therefore constitutional. When humans and AI agents jointly participate in organizational cognition and action, the organization must define what counts as an actor, which roles actors may occupy, where authority originates, how authority is delegated, what discretion is allowed, how action is attributed, how representations preserve traceability, how adaptation occurs, and when change requires amendment to deeper commitments. Without such constitutional architecture, organizations may increase automation while weakening accountability, accelerate local adaptation while fragmenting authority, or create extensive data traces without meaningful governance observability.

This paper asks: How can organizations be architected so that humans and AI agents can participate in distributed cognition, decision-making, execution, governance, and learning while preserving legitimacy, accountability, coherence, continuity, and adaptive capacity?

I develop a theory of Constitutional Organizational Architecture. The theory treats an AI-native organization as an adaptive enterprise governed by enduring but revisable principles, authority rules, role definitions, representation rules, adaptation procedures, and repair mechanisms. Constitutional Organizational Architecture is not a legal constitution, a policy manual, a software authorization model, or an enterprise architecture maturity score. It is a higher-order, formative configuration explaining how a small set of constitutional dimensions jointly constitute legitimate, accountable, observable, adaptable, and repairable mixed human-AI organizational action.

The argument proceeds in five steps. First, I synthesize relevant literatures around the problem of constituting mixed human-AI organizational action. Second, I identify limits in existing organizing paradigms without treating them as obsolete. Third, I define the foundational concepts needed to distinguish actors, roles, authority, permissions, delegation, accountability, adaptation, and constitutional evolution. Fourth, I present Constitutional Organizational Architecture and its central mechanisms. Fifth, I develop twelve propositions, boundary conditions, adverse effects, and failure modes that define the paper's theoretical model and research agenda.

The paper makes three contributions. First, it reframes AI-native organization as a constitutional organizing problem: the central question is how mixed human-AI action becomes authorized, attributable, bounded, observable, contestable, and repairable. Second, it develops Constitutional Organizational Architecture as a focal construct linking constitutional foundation, authority architecture, role architecture, representation architecture, adaptation architecture, and repair. Third, it advances an authority-centered account of AI governance by distinguishing technical autonomy from organizational authority and by specifying boundary conditions, tradeoffs, and failure modes under which constitutional architecture enables or constrains adaptation.