AI is not a single force: different kinds of systems—predictive, generative, agentic and embodied—alter expertise, authority and coordination in distinct ways; management theory must specify AI types to understand organizational impact.

Main Finding

The paper argues that treating “AI” as a single, undifferentiated technology obscures important differences between predictive, generative, agentic, and embodied systems. Precise specification of AI types is necessary because different modes of AI reshape organizational constructs (expertise, judgement, coordination, authority, institutional adaptation) in distinct ways. The authors offer a heuristic framework to differentiate contemporary AI systems and set a research agenda focused on shifting loci of agency, new decision architectures, and normative/institutional challenges.

Key Points

• AI is not monolithic: predictive (e.g., forecasting), generative (e.g., content creation), agentic (autonomous actors), and embodied (robots, IoT) systems operate on different logics and produce different organizational outcomes.
• Conceptual flattening—treating AI as a single construct—limits cumulative theorizing in management and organization studies.
• Different AI modes alter core organizational constructs:
  • Expertise: who/what is considered knowledgeable changes (human vs. system epistemics).
  • Judgement: distribution of discretionary decision-making shifts.
  • Coordination: coordination costs and mechanisms are reconfigured by AI affordances.
  • Authority: authority and oversight structures are renegotiated as systems act autonomously.
  • Institutional adaptation: organizational and field-level rules, norms, and accountability evolve.
• The paper presents a heuristic typology that clarifies the affordances of different AI systems and how they interact with organizational processes.
• A research agenda is proposed emphasizing the loci of agency, new decision architectures, and normative and institutional governance of increasingly powerful AI.

Data & Methods

• The study is conceptual and theoretical rather than empirical.
• Methods include:
  • Synthesis of existing literature across management, organization studies, and AI research.
  • Critical interrogation of prevailing conceptualizations that aggregate diverse AI systems.
  • Development of a heuristic typology differentiating predictive, generative, agentic, and embodied AI by their affordances and organizational effects.
  • Argumentative mapping of how each AI mode plausibly impacts organizational constructs and institutional dynamics.
• No primary quantitative dataset or empirical experiment is reported in the abstract; the contribution is analytical and agenda-setting.

Implications for AI Economics

• Disaggregate AI in empirical work: Economic models and empirical studies should distinguish between AI types because their effects on productivity, tasks, and markets differ (e.g., predictive systems augment forecasting tasks; agentic systems can substitute managerial decisions).
• Measurement and identification:
  • New metrics are needed to capture adoption and functional roles of different AI modes (not just “AI use” or software spend).
  • Empirical strategies should exploit variation in AI mode deployment (firm-level surveys, product/feature rollouts, platform data, natural experiments).
• Labor markets and tasks:
  • Task reallocation will vary by AI type—predictive AI may complement skilled workers, generative AI may substitute or augment creative tasks, agentic/embodied systems may automate coordination and routine supervisory roles.
  • Wage, employment, and skill-bias effects should be modeled conditional on AI mode and organizational decision architecture.
• Firm organization and market structure:
  • Agentic systems that internalize decision-making could alter firm boundaries, hierarchy, and transaction costs—implications for vertical integration and outsourcing.
  • Embodied and agentic AI could create scale economies and raise concentration risks; generative systems may lower content production costs, affecting entry and competition.
• Incentives, governance, and principal–agent problems:
  • New principal–agent problems emerge when decision rights shift to AI or hybrid human-AI teams; contract design and monitoring must adapt.
  • Regulatory and institutional responses (liability, accountability, certification) matter for diffusion and welfare.
• Macroeconomic and welfare considerations:
  • Aggregate productivity impacts will depend on complementarities between AI modes and human capital, adoption diffusion, and reallocation frictions.
  • Distributional effects and externalities (misinformation from generative models, systemic risks from agentic systems) require policy attention.
• Recommended empirical and modeling approaches for economists:
  • Microeconometric analyses using matched employer–employee, product, and platform data to identify heterogeneous treatment effects by AI type.
  • Field experiments and randomized rollouts within firms or platforms that vary the AI mode or decision architecture.
  • Structural models and general-equilibrium frameworks to capture task reallocation, firm entry/exit, and long-run distributional outcomes.
  • Agent-based and network models to study non-linear, systemic dynamics of agentic/embodied AI adoption.
  • Policy evaluation designs to assess governance interventions (liability rules, certification, disclosure requirements).

Taken together, the paper calls for richer conceptualization and empirical work that distinguishes AI modalities to accurately assess organizational and economic impacts, design incentives, and inform regulation.