Agentic AI turns HR tools into autonomous decision-makers, widening governance gaps; firms must adopt transparency, accountability, bias mitigation and enforce human oversight before deploying agentic systems in hiring, appraisal and workforce planning.

Main Finding

The paper argues that "agentic AI"—goal-directed, multi-step, adaptive AI agents capable of planning, persistent context, and tool orchestration—constitutes a qualitative shift for HRM versus traditional AI/ML. That shift creates novel governance imperatives (transparency, accountability, bias reduction, and human control). The authors develop a multi-theoretic conceptual framework (RBV, Institutional Theory, STS, AMO), seven testable propositions, and a proposed empirical design, and call for pre-deployment governance in critical HR functions—especially in emerging economies such as India.

Key Points

• Definition and distinct features of agentic AI in HRM:
  • High temporal autonomy, persistent memory, dynamic task decomposition and re-planning, tool/API orchestration, and in‑context adaptation.
  • Contrasts with traditional HR AI which is mostly stateless, single-task, and human-in-the-loop.
• HR functions most affected: end-to-end recruitment, onboarding, performance management, workforce planning, and learning & development.
• Primary governance risks:
  • Accountability gaps (diffuse responsibility among developers, deployers, users).
  • Opacity and explainability failures.
  • Multiple bias types (historical, representation, measurement, aggregation, evaluation).
  • Worker de-personalization and threats to meaningful work.
• Four governance imperatives proposed: transparency, accountability, bias mitigation, and retention of meaningful human control.
• Theoretical framing:
  • Resource-Based View (RBV): agentic AI + governance capabilities can be VRIN resources.
  • Institutional Theory: adoption and governance shaped by coercive, normative, mimetic pressures.
  • Socio-Technical Systems (STS) & AMO (Ability‑Motivation‑Opportunity) theories: emphasize design of human-AI systems and HR practice effectiveness.
• Contributions:
  • Conceptualization of agentic AI in HRM.
  • Multi-theoretic governance framework and seven theoretical propositions.
  • Managerial call to establish governance before deploying agentic agents in critical HR processes.
• Limitations: purely conceptual/systematic-review study; propositions not empirically validated; potential narrowness given rapidly evolving agentic-AI developments and focus on emerging economies.

Data & Methods

• Methodology: systematic conceptual review and literature synthesis across AI, HRM, organizational governance, and AI ethics literatures.
• Analytical tools:
  • Comparative construct (Table) contrasting traditional AI vs agentic AI across autonomy, decision scope, planning, memory, tool use, adaptability, and governance risk.
  • Selected literature synthesis (table) mapping key prior studies, findings, and gaps.
  • Multi‑theory integration to derive seven research propositions linking agentic AI capabilities, governance mechanisms, institutional context, and organizational performance.
• Empirical stance: proposes an empirical research design for future validation (no primary data collected or analyzed in this paper).

Implications for AI Economics

• Firm-level returns and strategic positioning:
  • Agentic AI plus robust governance can become a VRIN capability, raising returns to firms that invest in data infrastructure, governance, and human‑AI integration.
  • High upfront fixed investment and complementary governance skills may increase first‑mover advantages and barriers to entry.
• Labor market and distributional effects:
  • Greater automation of multi-step HR tasks implies potential substitution of HR routine and semi-skilled roles, altering HR labor demand and skill composition.
  • Risks of reinforcing historical biases can produce adverse distributional outcomes (unequal hiring, promotion), with macroeconomic consequences for inequality and human capital deployment.
• Regulatory and institutional economics:
  • Weak or lagging regulation in emerging economies heightens systemic risks; institutional pressures (multinationals, professional bodies) will shape adoption and de facto governance.
  • Governance can be framed as a costly public/private good; policy interventions (audit standards, disclosure rules, liability allocation) affect adoption incentives and social welfare.
• Market structure and competition:
  • Orchestration-capable agents that integrate external tools/APIs may increase platformization and vendor lock-in, concentrating market power among a few AI providers or ecosystems.
• Transaction costs and firm decision-making:
  • Effective governance reduces informational asymmetries and liability/ reputational risks, lowering transaction costs of deploying agentic systems; firms must weigh these governance costs against productivity gains.
• Research & measurement needs for AI economics:
  • Empirical estimation of productivity gains vs. governance and compliance costs; causal identification of labor displacement and wage effects; heterogeneity by firm size, sector, and country.
  • Evaluation of externalities (bias spillovers across firms), and the effects of different regulatory regimes on adoption, competition, and social welfare.
• Policy recommendations (high level):
  • Encourage measurable governance standards (transparency, auditability, human-in-the-loop thresholds) that reduce negative externalities while allowing productive use.
  • Support investments in data infrastructure and workforce reskilling to capture agentic AI benefits without exacerbating inequality.
  • Promote independent audits and liability clarity to align private incentives with social welfare.

Suggested next empirical steps for AI economists: quantify firm-level productivity effects of agentic HR agents, estimate adoption determinants (including governance costs), measure distributional labor impacts, and evaluate policy interventions (mandated audits, disclosure, liability rules) through quasi-experimental or randomized approaches.