Agentic AI in Business Process Management: A Critical Sociological Analysis of Capital, Isomorphism, and World-System Dynamics (2025)

OUS Academy in Switzerland
Sep 22
9 min read

Author: Maria Chen

Affiliation: Independent Researcher

Keywords: agentic artificial intelligence, business process management, organizational change, Bourdieu, institutional isomorphism, world-systems theory, governance, ethics, automation

Abstract

Agentic artificial intelligence—AI systems that plan, reason, and act with minimal human intervention—has moved from laboratory concept to practical deployment in 2025, with profound implications for Business Process Management (BPM). While early discourse focuses on efficiency and automation, this paper argues that the adoption of agentic AI is best understood as a socio-technical transformation that reconfigures fields of power, redistributes forms of capital, and reorganizes relations across the global political economy. Using Bourdieu’s concept of capital, DiMaggio and Powell’s institutional isomorphism, and Wallerstein’s world-systems theory, I develop a multi-level framework for analyzing how agentic AI reshapes organizations and markets. I then articulate a governance model for responsible deployment that balances innovation with fairness, transparency, and human agency. The paper concludes with research and managerial implications, noting that the winners of the “agentic turn” will be those able to convert technological capabilities into durable economic, social, and symbolic capital while aligning with evolving norms and core–periphery dynamics.

1. Introduction: From Tools to Actors

Artificial intelligence in organizations has traditionally occupied a supportive role—analyzing data, predicting outcomes, and assisting human decision-makers. In 2025, a shift is underway: AI systems increasingly exhibit agentic properties, chaining multi-step tasks, coordinating with enterprise applications, and initiating actions under defined policies. This transition alters BPM’s central assumptions about who performs work, how processes adapt, and what types of control and accountability are feasible.

The managerial promise is straightforward: faster cycle times, fewer errors, and responsive, data-driven operations. But the deeper significance lies elsewhere. Agentic AI introduces a new class of organizational actors. Agents do not simply execute instructions; they prioritize, negotiate constraints, and escalate exceptions. Their presence challenges role boundaries, redistributes expertise, and creates new arenas of contestation over legitimacy, trust, and value. Such changes cannot be captured by cost-benefit analyses alone. They must be situated within broader theories of power, institutional change, and the world-economy.

This article advances that broader view. I synthesize management practice with critical sociology, proposing a framework that helps leaders see beyond automation rhetoric to the structural consequences of agentic BPM.

2. Literature and Theoretical Framing

2.1 Agentic AI and BPM

Agentic AI is an umbrella term for AI systems capable of planning and executing sequences of actions toward specified goals. In BPM, these agents can monitor processes, trigger workflows, allocate resources, and coordinate with humans and other agents. Unlike traditional automation, agentic systems can adapt policies and execution paths to context, operating as semi-autonomous process participants. Recent scholarship has begun to outline design questions for agent governance, human-in-the-loop oversight, and safe rollback mechanisms. BPM research, which historically emphasized modeling, monitoring, and continuous improvement, now confronts the challenge of modeling actors that may themselves redesign parts of the process in real time.

2.2 Bourdieu’s Concept of Capital

Bourdieu identifies multiple, convertible forms of capital—economic (financial resources), cultural (knowledge, credentials, competencies), social (networks and relationships), and symbolic (recognized prestige and legitimacy). Organizations accumulate and mobilize these capitals within fields structured by competition and rules of recognition. The agentic turn shifts how capital is produced and recognized: who holds process knowledge, what counts as credible expertise, and which signals confer legitimacy to automated decisions.

2.3 Institutional Isomorphism

DiMaggio and Powell describe three mechanisms—coercive, mimetic, and normative—through which organizations grow more alike over time. Regulation (coercive), uncertainty-driven imitation (mimetic), and professional norms (normative) push firms toward similar structures and practices. In the agentic context, isomorphic pressures manifest as “must-have” governance patterns, audit controls, and disclosure norms for AI decisions. Vendors and consultants codify best practices that diffuse rapidly, while professional associations shape ethical expectations for human oversight.

2.4 World-Systems Theory

Wallerstein’s world-systems analysis locates firms and states in a core–semi-periphery–periphery structure mediated by flows of capital, technology, and labor. Agentic AI may reinforce core dominance if knowledge, data, and compute centralize in core economies; conversely, it may enable latecomer advantages if accessible platforms allow peripheral firms to leapfrog capability gaps. The distribution of benefits will depend on data sovereignty, access to compute, localization of models, and the ability of semi-periphery actors to build specialized niches.

3. Research Questions and Conceptual Model

RQ1: How does agentic AI reconfigure the distribution of economic, cultural, social, and symbolic capital within and across organizations?RQ2: Through which isomorphic mechanisms will agentic BPM practices stabilize into recognizable governance templates?RQ3: How will agentic BPM reshape core–periphery dynamics in the world-economy, especially regarding data, talent, and infrastructure?RQ4: What governance model best aligns innovation incentives with human agency, fairness, and accountability?

Conceptual model. I propose a three-level model:

Micro (task/role): Agents as co-workers; delegation boundaries; human–agent collaboration.
Meso (organizational field): Industry templates, compliance norms, and vendor ecosystems producing isomorphic convergence.
Macro (world-system): Cross-border data flows, compute concentration, and talent migration shaping competitive hierarchies.

Capital conversion (Bourdieu) operates at each level; isomorphic pressures stabilize patterns at the meso-level; world-system structures channel the distribution of gains and risks.

4. Methodological Approach

This is a theory-building, integrative review and conceptual analysis. I synthesize recent scientific and practitioner literature on agentic AI and BPM, classic sociological theory, and current managerial practice. The aim is not to test hypotheses statistically but to propose constructs, mechanisms, and propositions for future empirical research. Where illustrative examples are used, they serve as heuristic vignettes rather than claims about any single firm.

5. Analysis

5.1 Micro-Level Dynamics: Tasks, Roles, and Habitus

From tasks to trajectories. In traditional BPM, tasks are atomic and human-executed. In agentic BPM, tasks become parts of agent-managed trajectories that adapt to context. An order-to-cash flow, for example, may be continuously re-routed by an agent that predicts risk, schedules approvals, and contacts customers. The human role shifts from doer to supervisor, with selective intervention.

Habitus and skill conversion. Workers’ habitus—their embodied dispositions and tacit process knowledge—changes as routine tasks migrate to agents. Cultural capital is revalued: skills in prompt orchestration, exception handling, and model-risk literacy gain prestige; purely procedural experience may depreciate. This revaluation can produce status anxiety and resistance. Training that converts incumbent cultural capital (domain know-how) into agent-adjacent capital (policy writing, guardrail design) softens the shock.

Symbolic authority and trust. Who confers legitimacy on an agent’s decision? In early stages, legitimacy is borrowed from expert sponsors and formal governance bodies. Over time, symbolic capital accrues to systems that demonstrate reliability, fairness, and auditability. Transparent explanations, calibrated confidence, and consistent escalation behavior are crucial for building this capital.

Proposition 1. In agentic BPM, employees who can translate domain knowledge into governance policies will accumulate cultural and symbolic capital, becoming new organizational elites.

5.2 Meso-Level Dynamics: Isomorphism and Field Formation

Coercive pressures. Emerging regulations and audits create coercive isomorphism: organizations converge on similar documentation (model cards, data lineage), controls (segregation of duties for agents), and accountability schemas (who signs off on agent behavior). Procurement teams increasingly require third-party attestations for safety, privacy, and bias mitigation.

Mimetic pressures. Under uncertainty, firms imitate perceived leaders. This produces convergence on architectures (agentic orchestrators, action frameworks), metrics (time-to-resolution, autonomous action rate), and patterns (human-in-the-loop at high-impact steps). Mimetic adoption accelerates when vendors embed “reference blueprints” into platforms.

Normative pressures. Professional communities codify ethics guidelines, postmortem practices, and incident taxonomy. Normative isomorphism leads to shared language: “capability tiering,” “safety cases,” “replay sandboxes,” and “recovery time objective for agents.”

Proposition 2. Where regulatory salience is high, coercive isomorphism dominates; where competitive uncertainty is high, mimetic isomorphism dominates; over time, normative isomorphism consolidates and stabilizes governance templates.

5.3 Macro-Level Dynamics: World-System Reconfiguration

Data gravity and compute centralization. Core economies with hyperscale compute and rich enterprise data pull the value chain toward themselves. This threatens to widen capability gaps. However, semi-periphery regions can specialize: localized agents for language-specific customer operations, sovereign data zones, or industry-specific safety tooling.

Talent circulation. Agentic BPM elevates roles in model governance, safety engineering, and process cognition. Core economies initially attract this talent, but remote collaboration and platformization allow distributed centers of excellence to emerge in the semi-periphery.

Standards and legitimacy. Symbolic capital at the world-system level takes the form of standards, benchmarks, and certifications. Control over these confers agenda-setting power. If standards embed the realities of core contexts, periphery actors may be disadvantaged. Inclusive, multi-regional standard-setting reduces this risk.

Proposition 3. The net effect of agentic BPM on the semi-periphery depends on (a) access to compute at competitive prices, (b) localization of training data, and (c) participation in standard-setting processes that translate symbolic capital into recognized legitimacy.

5.4 Economic Capital and the “Agent Dividend”

Agentic AI promises a productivity “dividend” through reduced handling time, lower error rates, and higher throughput. Realizing this dividend requires organizational complementarity: redesigned processes, reallocated roles, and incentive structures that reward human–agent collaboration rather than output volume alone. Without such complementarity, firms risk “pilot purgatory,” where benefits remain local and non-scalable.

Proposition 4. Economic gains from agentic BPM scale non-linearly with governance maturity; early investments in measurement, audit, and safe-fail patterns increase the slope of returns.

5.5 Social Capital and Network Effects

Agentic BPM thrives on integration—APIs, event streams, and shared ontologies. Organizations with rich internal networks (high social capital) can more rapidly embed agents across units. Externally, consortia and industry data spaces amplify value by defining interoperable schemas and reciprocal audit protocols, turning social capital into a public good that reduces coordination costs.

5.6 Symbolic Capital: Narratives, Signaling, Legitimacy

Managerial narratives around “autonomous enterprise” generate symbolic capital that can attract customers, partners, and talent. But symbolic capital is fragile. High-profile errors can delegitimize programs quickly. Credible signals include independent evaluations, transparent incident reporting, and evidence of equitable outcomes across customer segments.

5.7 Cultural Capital: New Competencies and Professionalization

Agentic BPM professionalizes new roles: policy engineers, agent-ops specialists, evaluators, and AI safety auditors. Credentialing programs and communities of practice constitute cultural capital markets. Over time, hiring practices formalize these credentials, institutionalizing the new expertise hierarchy.

6. A Governance Model for Agentic BPM

6.1 Principles

Purpose Alignment: Tie agent objectives to explicit business and social goals.
Human Primacy: Preserve human override and clear escalation, especially for high-impact decisions.
Explainability and Auditability: Provide reasoned explanations, logs, and replayable traces.
Fairness and Inclusion: Measure outcomes across groups; remediate disparities.
Privacy and Security by Design: Minimize data collection; enforce least privilege.
Safe-Fail and Containment: Bound actions; implement rollbacks and kill-switches.
Continuous Learning with Guardrails: Update models with drift detection and change control.

6.2 Operating Model

Agent Lifecycle: Business case → policy design → data curation → simulation → pilot → staged rollout → post-deployment monitoring.
Three Lines of Defense: (1) Product teams own controls; (2) independent model-risk management tests assumptions; (3) audit and compliance attest to governance.
Metrics: Autonomy rate, intervention frequency, latency to escalation, fairness parity, explanation utility, and capital conversion ratio (how effectively agentic capability becomes economic, social, and symbolic capital).

6.3 Human–Agent Collaboration Patterns

Triage-and-Escalate: Agents handle routine work; escalate ambiguous cases with structured rationales.
Co-Drafting: Agents produce drafts; humans validate and add judgment.
Guardrail-Constrained Autonomy: Agents act within tight policies; humans review exceptions.
Committee of Agents: Diverse agents propose actions; a human arbiter selects or requests synthesis.

7. Propositions for Empirical Study

Capital Revaluation Hypothesis: Units investing in policy design and evaluation skills will experience upward mobility in internal status hierarchies.
Isomorphic Consolidation Hypothesis: Within three years of first adoption in a sector, governance templates will converge, reducing variance in control structures.
Core Leverage Hypothesis: Access to compute and proprietary data predicts share of agentic dividend captured by core-economy firms.
Symbolic Fragility Hypothesis: Publicly reported incidents have outsized effects on perceived legitimacy unless organizations demonstrate transparent remediation.

8. Managerial Playbook

Map Delegation Boundaries: Identify decisions suitable for agent autonomy; define red lines.
Invest in Cultural Capital: Upskill domain experts into policy authors and evaluators.
Institutionalize Governance: Create independent review boards and incident taxonomies.
Design for Equity: Build fairness metrics into success criteria; allocate remediation budgets.
Measure Symbolic Capital: Track trust indicators (NPS for agent interactions, explanation satisfaction).
Leverage Social Capital: Form cross-functional guilds and industry alliances for common schemas.
Plan for World-System Realities: Secure compute partnerships; negotiate data localization; diversify talent pipelines.

9. Ethical Considerations

Agentic BPM raises concerns about surveillance, deskilling, and unequal impact. A rights-respecting approach mandates data minimization, transparency, worker consultation, and meaningful mechanisms to contest automated decisions. Equity assessments should be routine, with clear accountability for remediating harms. Importantly, human agency is not the absence of automation but the ability to shape and override it.

10. Limitations and Future Research

This conceptual analysis synthesizes diverse literatures but lacks large-scale empirical testing. Future work should conduct cross-industry studies of agent governance, measure capital reallocation effects at the team level, and examine cross-national differences in regulatory and cultural receptivity. Comparative studies of core versus semi-periphery adoption patterns would illuminate world-system dynamics.

11. Conclusion

Agentic AI is changing BPM not merely by automating tasks, but by reconfiguring the distribution of power, expertise, and legitimacy within and across organizations. Viewed through Bourdieu, we see the revaluation of cultural and symbolic capital; through institutional isomorphism, we anticipate convergence toward shared governance templates; through world-systems theory, we grasp how global hierarchies may widen or narrow depending on access to compute, data, and standards. The decisive managerial challenge is to convert agentic capability into durable, just value—improving efficiency while reinforcing human primacy, fairness, and accountability. Organizations that learn to manage this socio-technical transformation will define the next era of process excellence.

References

Bourdieu, P. (1986). “The Forms of Capital.” In Handbook of Theory and Research for the Sociology of Education.
DiMaggio, P. J., & Powell, W. W. (1983). “The Iron Cage Revisited: Institutional Isomorphism and Collective Rationality in Organizational Fields.” American Sociological Review.
Wallerstein, I. (2004). World-Systems Analysis: An Introduction.
Hammer, M., & Champy, J. (1993). Reengineering the Corporation.
Davenport, T. H. (1993). Process Innovation: Reengineering Work through Information Technology.
Fettke, P. (2025). “Business Process Management and Artificial Intelligence: A Survey.” KI – Künstliche Intelligenz.
Vu, H., Klievtsova, N., Leopold, H., Rinderle-Ma, S., & Kampik, T. (2025). Agentic Business Process Management: Practitioner Perspectives on Agent Governance in Business Processes (arXiv preprint).
Sapkota, R. (2025). AI Agents vs. Agentic AI: A Conceptual Taxonomy and Research Agenda (arXiv preprint).
Nisa, U. (2025). “Agentic AI: The Age of Reasoning—A Review.” Scientific Discovery.