Theoretical Background

The paper draws on four broad streams of literature: organizing authority and coordination, organizing adaptation and learning, organizing human-technology relations, and organizing constitutive rules and legitimacy. These streams are complementary, but their assumptions become difficult to reconcile when AI agents participate in organizational action.

2.1 Organizing Authority and Coordination

Classical organization theory explains how organizations coordinate specialized work, allocate authority, and manage interdependence (Weber, 1947; March and Simon, 1958; Burns and Stalker, 1961; Chandler, 1962; Lawrence and Lorsch, 1967; Galbraith, 1973; Galbraith, 1974; Mintzberg, 1979; Nadler and Tushman, 1997; Puranam, 2018). Weber's account of bureaucracy emphasizes rule-bound authority and office-based legitimacy. March and Simon frame organizations as decision and communication systems. Burns and Stalker, Lawrence and Lorsch, Galbraith, Mintzberg, and later organization-design theorists explain how structure, differentiation, integration, and information-processing capacity should vary with task and environmental conditions. Contemporary microstructural approaches further decompose organizational design into problems of division of labor, task allocation, reward, and information provision.

This literature provides a foundation for understanding authority, coordination, role differentiation, and information-processing demands. Its implicit actor model, however, is mainly human and collective. Rules, structures, and routines may be formalized, but organizational action is ultimately attributed through human office holders, teams, or units. AI-native organizing complicates this assumption. If AI agents prepare decisions, trigger workflows, negotiate dependencies, or execute delegated steps, organizations must specify whether those agents occupy roles, how their authority is bounded, who remains accountable, and how their action is made contestable.

Enterprise architecture and operating-model literatures extend the analysis by connecting strategy, capabilities, processes, information, applications, technology, and governance (Bernard, 2012; Ross et al., 2006; Zachman, 1987; Lankhorst, 2017; The Open Group, 2022). They help organizations reason across domains rather than treating structure, process, and technology separately. Yet enterprise architecture usually represents organizational capabilities and systems rather than constituting legitimate organizational actors. It can show that an application supports a process or capability, but it does not by itself answer whether an AI agent has legitimate authority to act, how that authority was delegated, or when a topology change crosses into constitutional change.

2.2 Organizing Adaptation and Learning

Dynamic capabilities, organizational learning, organizational memory, cybernetics, viable systems, and complex adaptive systems explain how organizations sense, learn, adapt, and survive changing conditions (Ashby, 1956; Beer, 1972; Beer, 1979; Bertalanffy, 1968; Holland, 1995; Kauffman, 1993; Simon, 1962; Argyris and Schon, 1978; Crossan et al., 1999; March, 1991; Senge, 1990; Walsh and Ungson, 1991; Teece et al., 1997; Teece, 2007). Dynamic capabilities emphasize sensing, seizing, and reconfiguring. Organizational learning and memory literatures examine how experience becomes retained, interpreted, and institutionalized. Cybernetics and viable-system theory highlight feedback, regulation, variety, recursion, and control. Complex adaptive systems theory foregrounds emergence, self-organization, modularity, and adaptation under interdependence.

These literatures are essential for AI-native organizations because AI agents increase both the speed of local action and the volume of observable traces. They suggest that adaptation depends on feedback, learning, and reconfiguration rather than static structure. They also warn that adaptation without coherent regulation may fragment the system. Yet they typically do not distinguish runtime adaptation from constitutional evolution with the precision needed for mixed human-AI organizing. An organization may adjust a workflow, reassign a role, change a model threshold, alter an approval rule, or amend a principle. These changes are not equivalent. Some are operational adjustments; others revise the authority basis under which future action becomes legitimate. A constitutional theory must therefore explain both how adaptation occurs and which commitments must remain stable, reviewable, and amendable.

2.3 Organizing Human-Technology Relations

Sociotechnical systems, digital transformation, platform organization, AI governance, human-AI interaction, algorithmic organizing, and multi-agent systems literatures focus attention on the interaction between humans, technologies, tasks, and institutional settings (Emery and Trist, 1960; Trist and Bamforth, 1951; Bharadwaj et al., 2013; Vial, 2019; Gawer, 2014; Parker et al., 2016; Nambisan et al., 2017; Faraj et al., 2018; Amershi et al., 2019; Crawford, 2021; Floridi and Cowls, 2019; Jobin et al., 2019; Russell and Norvig, 2021; Shneiderman, 2022; Wooldridge, 2009). Sociotechnical theory establishes that technical and social systems should be jointly designed. Digital transformation research explains how digital technologies reshape strategy, innovation, and organizational forms. AI governance and responsible AI literatures articulate principles such as transparency, accountability, human oversight, fairness, safety, and contestability. Human-AI interaction research offers design guidance for human use of AI systems. Multi-agent systems research provides formal tools for modeling computational agents, coordination, protocols, and norms.

These literatures make clear that AI systems are not neutral tools placed outside organizational life. They participate in sociotechnical arrangements and may shape the flow of work, knowledge, and decision-making. Still, the organizational question remains underdeveloped. Human-AI collaboration research often focuses on task interaction. Responsible AI often focuses on model development and use. Multi-agent systems can define computational protocols and norms but does not automatically settle organizational legitimacy. The organization must decide which agents are admitted, what role they occupy, what jurisdiction they have, how they can be supervised, and how action is attributed when computational execution becomes organizational action.

2.4 Organizing Constitutive Rules and Legitimacy

Institutional theory, governance theory, and constitutional analogies help explain why rules, roles, authority, and legitimacy are not merely technical arrangements (Meyer and Rowan, 1977; DiMaggio and Powell, 1983; Scott, 2014; Ouchi, 1980; Williamson, 1985; Freeman and Hannan, 1983). Institutions define socially recognized rules and meanings. Governance arrangements allocate decision rights and accountability. Constitutional concepts distinguish ordinary action from the rules that constitute legitimate action. These literatures clarify that organizations do not only coordinate work; they establish the conditions under which action is recognized as appropriate, accountable, and binding.

This distinction is central to the paper's use of constitutional language. A constitution is not treated here as a legal transplant or as a metaphor for "important rules." It is treated analytically as the rule-order that defines who may act, under what authority, with what accountability, through what representations, and by which amendment or repair procedures. Constitutional rules are therefore constitutive rather than merely regulatory: they do not only constrain action after the fact; they establish the conditions under which organizational action can count as legitimate. Legitimacy is not reducible to documentation or traceability. Traceability can support legitimacy, but legitimacy also depends on accepted authority sources, institutional context, contestability, and the capacity to resolve disagreement.

For AI-native organizing, the constitutional lens is useful but must be used carefully. The point is not that firms are states, that every organization requires a legal constitution, or that AI systems should become citizens. The point is that organizations need an architectural layer of constitutive commitments: principles, authority sources, role rules, representation requirements, delegation conditions, amendment pathways, and repair mechanisms that make distributed action legitimate and governable. The constitutional framing therefore complements organization design, enterprise architecture, AI governance, and adaptive-systems theory by asking how the organization's own action system is constituted.

Table 1 summarizes the differentiation that motivates this paper.