Agentic AI is shifting where decision authority resides inside firms, creating a new 'agentic authority' rooted in data and performance; leaders will need to act as 'authority integrators' who reconcile AI recommendations with accountability and organisational values.

Main Finding

The article introduces and theorises "agentic authority"—a form of organisational authority that derives legitimacy from AI systems (data, predictive performance, system design) rather than from formal position or interpersonal expertise. Agentic authority redistributes decision influence across socio-technical systems and requires a new leadership role, the "authority integrator", who mediates between AI-driven prescriptions and normative, accountability, and legitimacy concerns. The shift produces hybrid authority structures, new accountability challenges, and reconfigured organisational power both inside and beyond firm boundaries.

Key Points

• Definition
  • Agentic authority: organisational influence granted to AI systems because of perceived objectivity, procedural fairness and demonstrated performance, not because they occupy formal roles or human expert status.
• Distinction
  • Moves beyond algorithmic decision-support to agentic AI (auto-pilot end of autonomy continuum) that plans, executes multi-step workflows and adapts with limited human intervention.
• Mechanisms producing agentic authority (Figure 1)
• Datafication: organisational phenomena are translated into quantifiable data that privilege modelable problems.
• Performance feedback: repeated superior outputs strengthen credibility and reliance.
• Institutionalisation: embedding AI outputs into routines, governance and SOPs normalises their authority.
• Cognitive delegation: managers defer judgement under complexity, time pressure and accountability burdens.
• Legitimacy dimensions for algorithmic systems
  • Epistemic legitimacy (accuracy/objectivity), procedural legitimacy (consistency/fairness), and performance legitimacy (demonstrable effectiveness).
• Leadership implications
  • Leadership becomes hybrid: human judgement + machine recommendation.
  • New role: authority integrator — responsible for reconciling analytic outputs with ethics, stakeholder expectations, organisational values and legal accountability.
  • Tension between opacity and trust: black-box systems can be trusted instrumentally while being hard to contest.
• Redistribution, not replacement
  • Agentic authority redistributes authority across managers, technical experts, AI systems, vendors and infrastructures—sometimes shifting decision influence outside firm boundaries (vendor lock-in, platform dependency).
• Context sensitivity
  • Strongest in data-rich, structured, measurable domains (pricing, trading, credit scoring, maintenance), but expanding toward strategic and less-structured domains as capabilities grow.

Data & Methods

• Approach: conceptual/theoretical paper built on literature synthesis and framework development.
  • Draws on organisational theory (Weber; French & Raven), human-AI collaboration literature, digital leadership scholarship, and recent AI/agentic AI work.
  • Integrates insights from studies on algorithmic legitimacy, automation bias, institutionalisation and socio-technical systems.
• Output: a conceptual framework (including the four mechanisms) and the authority integrator construct to analyze how AI-mediated authority emerges and interacts with managerial authority.
• Empirical content: no primary quantitative data or novel empirical tests; illustrative examples and references to prior empirical findings (e.g., domains where AI already autonomously executes decisions).

Implications for AI Economics

• Firm decision rights and governance
  • Allocation of decision rights will shift from humans to algorithms in measurable ways; economic models should incorporate endogenous assignment of decision authority to AI.
  • New principal–agent problems: agents (managers) may delegate cognitive tasks to AI, but accountability and incentives need to be redesigned.
• Labour markets and skill demand
  • Rising demand for "authority integrators": skills in governance, ethics, explainability, arbitration between model outputs and normative constraints.
  • Possible decline in some middle-management decision roles for structured tasks, offset by growth in oversight, compliance and AI-integration jobs.
• Productivity and value of data
  • Agentic authority amplifies returns to high-quality data, infrastructure and model performance—strengthening scale and scope economies for data-rich firms.
  • Empirical implications: measure productivity gains from shifting decisions to agentic AI, but account for fragility/risk externalities (e.g., systemic failures).
• Market structure and concentration
  • Vendor/platform dependencies can move decision influence outside firms, increasing market power for top model/infrastructure providers; leads to potential entry barriers and lock-in effects.
  • Economists should evaluate how value accrues between firms using AI and AI vendors, and implications for competition policy.
• Risk, externalities and insurance
  • Opacity and institutionalisation increase systemic risk and the cost of failures; new markets for AI auditing, model insurance, and third-party assurance may grow.
  • Regulatory design (liability rules, mandatory logging/explainability, auditability) will affect adoption incentives and incidence of agentic authority.
• Measurement and empirical research agenda
  • Proposed measures: share of decisions delegated to agentic systems; an "agentic authority index" combining datafication, institutional embedding, performance reliance and cognitive delegation; vendor dependency metrics; incidence of contested/failed AI decisions.
  • Research questions: How does agentic authority affect firm value, risk exposure, investment in data vs. human capital? Does agentic authority raise or lower aggregate welfare when accounting for efficiency gains and contestability losses? What contract forms optimally allocate residual control and liability when decisions are partly automated?
• Policy and regulatory implications
  • Need to design accountability regimes that balance innovation incentives with contestability, transparency, and externalities mitigation.
  • Competition policy should monitor concentration in AI provisioning and cross-firm dependencies that transfer internal decision authority externally.

Short practical takeaways for economists and policy analysts: - Model decision-making as a hybrid of human and agentic actors, with authority endogenous to performance, institutional embedding and data infrastructure. - Track organizational metrics (delegation rates, embedding of AI in SOPs, vendor reliance) to empirically study economic impacts. - Anticipate labour reallocation toward governance, auditing and integrative roles and consider policy responses (retraining, certification, liability regimes).