Submitted:
23 October 2026
Posted:
26 October 2026
You are already at the latest version
Abstract
Leadership is increasingly delivered, in part, by software. Agentic artificial intelligence (AI) systems now monitor team activity, coordinate work across people and tools, schedule and assign tasks, deliver feedback and performance nudges, and broker information between leaders and followers — functions that organisations once obtained only from human managers. This has prompted the suggestion that leadership itself is becoming a service: a set of functions that can be specified, subscribed to and supplied by systems rather than embodied in persons. This conceptual paper examines that suggestion critically. Drawing on functional leadership theory, the automation–augmentation paradox, and the literature on algorithmic management and trust in AI, I develop a function-by-delegability framework that classifies five leadership functions — sensing and monitoring, coordination, deciding and allocating, developing and giving feedback, and representing and meaning-making — according to how far each can responsibly be delegated to agentic systems. I argue that delegability is governed less by technical capability than by accountability: I introduce the accountability floor, the principle that functions whose delegation would break the chain of accountability may be automated in execution but never in ownership. I identify four risks of delegation — legitimacy and follower trust, bias at scale, accountability gaps, and erosion of the leadership pipeline — propose governance and design principles, and derive three propositions for empirical research.
Keywords:
agentic AI
; leadership
; leadership as a service
; algorithmic management
; delegation
; accountability
; automation–augmentation paradox
1. Introduction
Much of what organisations call leadership is, on inspection, a bundle of recurring functions: watching the environment and the team’s work, coordinating interdependent activity, allocating tasks and resources, developing people through feedback, and representing the group—to itself and to others—by explaining what its work means. For most of organisational history these functions could only be performed by persons, because they required continuous attention, judgment across incomplete information, and a presence that others could trust, question and hold responsible. Agentic AI systems now perform versions of all five. Software monitors dashboards and communication channels and flags exceptions; assigns and reassigns tasks as conditions change; schedules, reminds and nudges; drafts and in some settings delivers feedback on individual performance; and summarises, routes and brokers information between senior leaders and the front line—in some organisations at a scale and continuity no human manager could match.
This invites a provocative formulation: leadership as a service. If leadership functions can be specified as service levels, delivered by subscription platforms and measured by output metrics, organisations may increasingly buy leadership the way they buy computing—as a service layer rather than a hierarchy of people. The formulation clarifies what is at stake: a service can be delegated; a responsibility cannot so easily be transferred. When an agentic system monitors a team, coordinates its work and nudges its members, who is leading? And when it allocates work unfairly, gives damaging feedback, or coordinates the team efficiently in the wrong direction, who answers for it?
This paper addresses three questions. First, which leadership functions can be delegated to agentic AI, which can be augmented by it while remaining human-led, and which must remain human in more than name? Second, what risks arise as delegation deepens, and for whom? Third, what governance keeps delegated leadership legitimate and accountable? My answer, developed below, is that the binding constraint on delegating leadership is not what AI systems can do but what organisations can still account for. I call this constraint the accountability floor, and I argue that beneath it, leadership may be delivered through AI systems, but it cannot be delivered by them in the sense that matters for legitimacy: there remains a named human who owns the function and answers for its exercise.
The paper is conceptual. It reports no new data and advances no empirical claims; its contribution is a framework, a set of design principles, and propositions offered for testing. Section 2 sets out foundations. Section 3 presents the function-by-delegability framework. Section 4 and Section 5 address risks and governance, Section 6 states three propositions, and Section 7 concludes with limitations.
2. Foundations
Leadership as functions, not persons. Functional leadership theory holds that leadership is best understood not as the traits of individuals but as a set of functions a group needs performed—monitoring, coordinating, developing, representing—whoever or whatever performs them (Morgeson, DeRue & Karam, 2010). This lens is what makes the present question coherent: if leadership is functions, then asking whether a function can be delegated to a system is a well-formed question, and different functions can receive different answers. It also warns against the opposite error: a function performed is not necessarily leadership achieved, because groups also need functions—legitimacy, meaning, responsibility—that are constituted by who performs them and to whom that performer answers.
The automation–augmentation paradox. Raisch and Krakowski (2021) show that automation and augmentation are interdependent over time: automating a task changes what humans must do well, and augmenting humans changes what can next be automated. Applied to leadership, the paradox cuts deep. Delegating routine leadership functions (monitoring, scheduling) can free human leaders for judgment and development work—or it can remove exactly the daily contact through which leaders learn their teams and earn the trust that makes their judgment acceptable. Which path unfolds is a design and governance outcome, not a technological given. Work on competitive advantage adds a related point: AI adoption creates new human–machine capabilities rather than simply upgrading old ones, and those new capabilities can be unrelated to—even negatively related to—traditional ones (Krakowski, Luger & Raisch, 2023). The leadership capabilities an organisation needs when AI performs part of leadership are likewise new: specifying functions, supervising systems, auditing outputs and intervening well.
Algorithmic management and trust. The closest empirical tradition is algorithmic management: algorithms directing, evaluating and disciplining workers, documented extensively in platform work and increasingly in conventional organisations. Kellogg, Valentine and Christin (2020) organise this literature around how algorithms control work—restricting and recommending, recording and rating, replacing and rewarding—and how workers resist and adapt. Two lessons transfer. First, algorithmic direction is experienced as authority whether or not it is labelled leadership, so legitimacy questions apply in full. Second, control through opaque systems generates distinctive harms—contested evaluations, gaming, and a sense of being managed by no one in particular. Trust in AI, moreover, is built differently from trust in people: it responds to transparency and reliability, and is fragile in different ways (Glikson & Woolley, 2020). The right question is therefore rarely “AI or human?” but “which structure of human and AI decision rights, for which class of decision, under what safeguards?” (Dwivedi et al., 2021; Shrestha, Ben-Menahem & von Krogh, 2019).
3. A Function-by-Delegability Framework
I assess each leadership function on two dimensions: technical delegability—how reliably current and near-term agentic systems can execute it—and accountability criticality—how much harm follows, and how hard remedy is, when the function is exercised badly and no human owns it. A function is a candidate for delegation only when delegability is high and accountability criticality is low, or when governance (Section 5) demonstrably carries the criticality.
Leadership function—What agentic AI does—Delegability—Reasoning
---—---—---—---
Sense / Monitor—Continuous monitoring of metrics, communications and environment; exception flagging; early-warning summaries—High—Execution is pattern recognition over defined signals; —errors are visible and correctable; a human reviewing exceptions retains real oversight. Main caution: what gets measured gets managed—the choice of signals remains a human leadership act.
Coordinate—Scheduling, task assignment and reassignment, workflow routing, meeting and deadline orchestration—High, within set parameters—Coordination against explicit rules and constraints is where agentic systems are strongest. Delegability falls as coordination embeds contested priorities; setting the parameters—whose work yields when resources conflict—is allocation, not coordination.
Decide / Allocate—Recommending or making allocations of budget, tasks, opportunities, evaluation outcomes—Partial: recommend yes, commit no—Systems can generate and rank allocation options with supporting evidence. But allocations distribute real goods and burdens among identifiable people; they require reasons that can be given to those people by someone answerable to them. High-stakes, irreversible or value-laden allocations must be committed by a named human.
Develop / Feedback—Performance data summaries, coaching prompts, draft feedback, nudges, learning recommendations—Partial: support yes, deliver with care—AI can prepare feedback well and nudge consistently. But developmental feedback works through a relationship: its effect depends on the follower believing the giver knows them and is accountable for the judgment. Fully automated delivery of consequential feedback risks both error and a distinct harm—being judged by a system no one will discuss it with. Routine, low-stakes feedback can be delegated; consequential feedback must be owned, and normally delivered, by a human.
Represent / Meaning-make—Explaining purpose, narrating change, representing the group to outsiders, embodying values in crisis—Low: do not delegate ownership—Meaning-making is constitutive, not informational. A message about why the work matters carries weight because of who stands behind it and what they stake on it. Systems can draft and distribute; they cannot represent, because representation requires a party who can be held to the meaning they made. In crises this is at its sharpest: followers look for a person who answers.
The accountability floor. Across the matrix a single boundary appears, which I call the accountability floor: a leadership function may be automated in its execution, but it may not be delegated in its ownership below the point at which a named human can explain it, justify it to those it affects, and bear consequences for it. Delegation that respects the floor uses AI to extend a leader’s reach. Delegation that breaches the floor produces leadership-shaped activity with no leader attached—output that functions like leadership until something goes wrong, at which point the organisation discovers that no one was, in fact, leading. The floor also explains why delegability is asymmetric: functions can move down (toward more human ownership) quickly and rightly after a failure, whereas moving them up (toward delegation) should require evidence and re-certification.
4. Risks
Legitimacy and follower trust. Leadership depends on followers accepting influence as rightful. Influence exercised by an opaque system may secure compliance while forfeiting legitimacy—and legitimacy deficits surface at the worst moments: change, crisis, sacrifice. Because trust in AI follows different rules from interpersonal trust (Glikson & Woolley, 2020), a delegated function can appear to work—high compliance, no complaints—while the relational capital leadership depends on quietly depletes.
Bias at scale. A biased human manager harms a team; a biased leadership system harms every team it touches, uniformly and continuously, and its consistency can make the bias harder to see. Allocations of tasks, visibility and opportunity shape careers; systematic skew in those allocations is discrimination at organisational scale, executed at machine speed.
Accountability gaps. When monitoring, coordination and recommendation are distributed across vendors, models and workflows, responsibility for outcomes becomes genuinely unclear—not merely evasively claimed to be. The accountability floor is breached not by decision but by architecture: no single human can reconstruct why a follower was allocated, rated or nudged as they were.
Leadership pipeline erosion. Junior and middle managers develop judgment by performing monitoring, coordination and feedback functions themselves. If those functions are delegated to systems, organisations consume their existing leadership stock without growing the next one—the automation–augmentation paradox in its developmental form (Raisch & Krakowski, 2021). The cost arrives years later, as a shortage of people capable of supervising the very systems that replaced their training ground.
5. Governance and Design Principles
- Named accountable leader for every delegated function. Each function performed by an agentic system has a named human owner, known to followers, who can explain, justify and reverse its outputs. Role titles are not sufficient; the name is the mechanism.
- Transparency to followers. Followers are told, plainly and in advance, when an AI system is monitoring, coordinating, evaluating or nudging them, what data it uses, and what it can and cannot decide. Covert algorithmic leadership is illegitimate by design, whatever its accuracy.
- Appeal and override routes that work. Any follower affected by a consequential AI-exercised function can reach a human with authority to review and overturn it, within a stated time. Overrides and appeals are logged and learned from; a rising override rate is a governance signal, not a nuisance.
- Pre-delegation evidence and periodic re-certification. A function is delegated only after the system has demonstrated reliability on that function in that context, and delegation lapses unless periodically re-certified by the accountable leader—including checks for bias, drift and follower experience. Delegation is a licence with a renewal date, not a permanent transfer.
6. Propositions
Proposition 1. Follower trust and perceived legitimacy are higher for AI-exercised leadership functions when a named accountable human leader is identified and appeal routes are known, than for identical functions exercised without them—and the difference is larger for consequential functions (allocation, feedback) than for routine ones (scheduling, monitoring).
Proposition 2. Organisations that delegate monitoring and coordination functions to agentic systems without a deliberate replacement development pathway will show declining internal promotion readiness for leadership roles over a three-to-five-year horizon, relative to organisations that redesign managerial development alongside delegation.
Proposition 3. Periodic human re-certification of delegated leadership functions reduces the incidence and duration of systematic allocation bias, compared with one-time pre-deployment validation, because model behaviour and organisational context drift jointly over time.
7. Limitations and Conclusions
This is a conceptual paper. Real leadership functions blend into one another, delegability will shift as systems improve, and context—sector, culture, regulation, workforce expectations—will move the accountability floor for particular functions. The propositions are stated for testing, not as findings. The “leadership as a service” formulation itself has limits: it captures what can be specified and supplied, while leadership plausibly contains a remainder—presence, example, responsibility borne in person—that the service metaphor cannot reach; that is part of my argument rather than a defect of it.
The question posed by agentic AI is not whether software will perform leadership functions; it already does. The question is whether organisations will delegate those functions above the accountability floor—with a named human who owns them, followers who know an AI is acting, routes of appeal that work, and licences that expire—or below it, where leadership activity proceeds with no one answerable for it. “Leadership as a service” is an acceptable description of the first arrangement and a warning about the second: a service can be cancelled, renegotiated and held to account by the people who buy it. Leadership exercised over people must meet at least that standard, and then a higher one—someone must stand behind it.
Author Contributions
Lalit Gautam: conceptualisation, framework development, writing—original draft, review and editing. The author is solely responsible for the content of this paper.
Conflicts of Interest
The author is affiliated with Sensegrass, as stated in the author affiliation. This research received no external funding, and Sensegrass had no role in the design, preparation or decision to publish this work. The author declares no other conflicts of interest.
Declaration
This preprint has not been submitted to, and is not under consideration by, any journal or conference. It is a distinct work from the author’s other papers in preparation, including the author’s systematic review of AI in strategic decision-making and their framework paper on agentic AI in management consulting, and shares no text with them.
References
- Dwivedi, Y. K., et al. (2021). Artificial intelligence (AI): Multidisciplinary perspectives on emerging challenges, opportunities, and agenda for research, practice and policy. International Journal of Information Management, 57, 102216. [CrossRef]
- Glikson, E., & Woolley, A. W. (2020). Human trust in artificial intelligence: Review of empirical research. Academy of Management Annals, 14(2), 627–660. [CrossRef]
- Kellogg, K. C., Valentine, M. A., & Christin, A. (2020). Algorithms at work: The new contested terrain of control. Academy of Management Annals, 14(1), 366–410. [CrossRef]
- Krakowski, S., Luger, J., & Raisch, S. (2023). Artificial intelligence and the changing sources of competitive advantage. Strategic Management Journal, 44(6), 1425–1452. [CrossRef]
- Morgeson, F. P., DeRue, D. S., & Karam, E. P. (2010). Leadership in teams: A functional approach to understanding leadership structures and processes. Journal of Management, 36(1), 5–39. [CrossRef]
- Raisch, S., & Krakowski, S. (2021). Artificial intelligence and management: The automation–augmentation paradox. Academy of Management Review, 46(1), 192–210. [CrossRef]
- Shrestha, Y. R., Ben-Menahem, S. M., & von Krogh, G. (2019). Organizational decision-making structures in the age of artificial intelligence. California Management Review, 61(4), 66–83. [CrossRef]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2026 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
Copyright: This open access article is published under a Creative Commons CC BY 4.0 license, which permit the free download, distribution, and reuse, provided that the author and preprint are cited in any reuse.