Guest editorial: The architecture of accountability: algorithmic autonomy and the discipline of stewardship

For decades, the competency paradigm has served as a dominant boundary condition in leadership theory. Predicated on the assumption that organizational outcomes are the determinative outcomes of personal agency and interpersonal dynamics, models such as transformational leadership (Bass, 1985) and leader–member exchange (LMX; Graen and Uhl-Bien, 1995) have tended to bracket a leader's interpersonal traits from the performance realities of the architectures leaders authorize. As a result, these frameworks can be structurally underspecified for the present moment: the rise of the algorithmic agent as a primary organizational actor whose effects are mediated through scale, opacity and automated optimization.Grounding our inquiry into sociotechnical systems (STS) theory (Trist, 1981), this volume treats leadership in the AI era as an emergent property of interaction between social actors and technical infrastructures. We argue that a traditional unit of analysis in organization development (OD), the leader's mindset, no longer suffices on its own when algorithmic architectures function as consequential organizational actors. Following Drucker's (1954) framing of management as a social organ responsible for institutional performance, we move beyond a primarily utilitarian inquiry into how leaders use AI to improve productivity and instead examine how organizational responsibility is operationalized, or surrendered, through the architectures that leaders authorize.The urgency of this critique is no longer theoretical; it is being adjudicated in real time (Grabenstein, 2026a, b). As seen in the 2026 landmark trial in Los Angeles, Meta's CEO was confronted with alleged systemic outcomes of Instagram's architecture, sharpening what this volume terms the performance gap between stated values and observable results (Tan, 2026). Drucker's “mirror test” offers a governance-relevant ethic for this gap: ethics requires that a leader's actions remain compatible with the person they can live with seeing “in the mirror” each morning (Drucker, 1999/2005). When leaders authorize systems of infinite reach yet represent the resulting harms as operationally intractable (too complex, too diffuse or too difficult to govern), the managerial posture shifts from stewardship to authorization without effective oversight, leaving a social organ unmanaged at scale.We anchor this inquiry in the “Kaplan Paradox”: a condition in which a leader's systemic liability expands in direct proportion to algorithmic reach, even as operational transparency, and thus perceived managerial control, diminishes. The paradox is sustained by shadow text: layers of algorithmic operation that are functionally inaccessible to the human agent. When leaders authorize opaque architectures through a digital “click-to-approve” autonomy, they define an automated reality while remaining legally and socially tethered to its outcomes.As Griffin (2025) demonstrates, this “black box” can mask the deliberate ethical agency of developers – agency that management has a professional duty to govern. In agency-theoretic terms, leaders remain principals even when decision rights are delegated to technical agents; delegation changes monitoring and incentives, not responsibility for outcomes (Eisenhardt, 1989). By failing to embed the ecosystems of responsibility advocated by Stahl (2023), leaders enter a governance mode that resembles an ethical exception (Agamben, 2005), in which normal expectations for explanation, auditability and duty of care are suspended in practice. Reliance on delegation that is not readily auditable constitutes a structural breach of the duty of care, rendering leadership responsible for systemic malfunction, even when leaders experience diminished operational visibility.The papers in this part provide the diagnostic and statutory components required to repair this professional breach.If Part I provides the diagnostic audit of the “Kaplan Paradox”, Part II turns to the rigorous technical labor of the active steward, managing not only people but also the integrity of the systems through which power is exercised within the organizational control room (Drucker, 2010). Within this frame, we explicitly reject the managerial sedative – the assumption that empathy or culture alone can resolve the Kaplan Paradox. Empathy without systemic maintenance is not governance; it is performance.Although Drucker did not live to see the AI era, his core claim remains germane: managers are responsible for designing institutions to produce specific results and for maintaining the legitimacy of those results through disciplined practice (Drucker, 1954, 1973). Extending this legacy, we shift the unit of analysis from a conventional human-in-the-loop (HIL) model to a machine-in-the-human-loop (MIL) architecture, in which algorithmic agents are treated as subordinate within a shell of human stewardship (Shneiderman, 2022), maintaining that shell requires more than leadership style. It requires algorithmic discretion as a digital evolution of street-level discretion: practical judgment applied to mitigate the rigidity and externalities of rule-bound systems under real-world conditions (Lipsky, 1980).In this control room view, the leader must serve as the conscience of the architecture. Refilling the managerial vacancy created by automated delegation requires three stewardship postures derived from Drucker's functionalist ethics:Mirror test. The mirror test operates as a hard constraint against the “I didn't know” defense: if a leader authorizes an agent that acts in ways the leader would not personally endorse, the leader has failed in self-management (Drucker, 1999/2005).Feedback analysis. Feedback analysis operationalizes accountability as a method: leaders must declare the system's intent, compare expected outcomes with actual results and intervene, via manual override, when the two diverge (Drucker, 1999/2005).Pioneering the future. For Drucker, management is an organ of society charged with making the future rather than inheriting it by default (Drucker, 1954, 1973). In this sense, the “don't poke the bear” posture is not prudence but abdication: leaders cannot be passengers in machines they built but refuse to drive (Woffard, 2022).When managers do not assume this architectural responsibility, a vacancy forms: a vacuum of integrity in which both steward and user are reduced to cogs within a system whose defaults become destiny (Trist, 1981). The papers in this part examine the mechanisms by which organizations can reclaim agency through the technical labor of algorithmic discretion, re-establishing the manual override as a professional duty rather than an optional intervention.Management discipline in the AI era is defined by the manual override. We define algorithmic discretion as the digital evolution of street-level discretion required to mitigate the inherent rigidity of automated rule sets.To govern the machine loop, the steward requires constant telemetry.Together, they signal precisely when the architecture requires an active override.The greatest vulnerability of the system architect is operational opacity. As Zuboff (2019) warns, the ultimate cost of algorithmic governance is the erosion of the right to the future tense, the human capacity to act with intent.Together, this two-volume collection moves beyond the traditional boundaries of the competency paradigm to establish a formal architecture of accountability. By integrating the diagnostic rigor of an institutional audit with the functional requirements of the active steward, these papers provide a theoretical and practical roadmap for reclaiming agency within machine-mediated organizations. This transition, from managing interpersonal behavior to governing authorization logic, is not merely a technical adjustment; it is a fundamental evolution of the leadership duty of care.The Kaplan Paradox reminds us that as algorithmic reach expands, the vacuum of integrity created by automated delegation can no longer be filled by managerial sedatives or performative ethics. Instead, it requires the rigorous, technical labor of algorithmic discretion and the operational reality of the manual override. Ultimately, the control room is defined throughout this special issue as not only a technical mechanism of absolute authority but also a site of professional responsibility requiring the discipline of practice, where the machine is brought back under the stewardship of the human hand and the institutional conscience.This article is based on secondary analysis of published accounts and does not involve new research with human participants; therefore, ethics approval was not required.No new data was created or analyzed in preparation of this work. Data sharing is not applicable to this article.During the preparation of this work, the author used ChatGBT to assist with citation verification and formatting. After using this tool, the author reviewed and edited the content and takes full responsibility for the content of the published article.This Special Issue is the result of a nine-month journey that began at the International Leadership Association (ILA) Virtual AI Summit 2025, following an editorial panel discussion I facilitated. I am deeply indebted to Stefanie Johnson for her early vision in appointing me as Guest Editor and for granting me the intellectual freedom to chart the architecture for this issue. I also thank Rickard Enstroem, who provided steady stewardship and vital continuity as he assumed the leadership of the Leadership and Organization Development Journal.A project of this complexity relies on an invisible infrastructure of professional labor. I wish to extend a special thanks to my friend, colleague, and fellow editor, Suzanne Joy Clark, whose assistance in co-auditing and managing the extensive volume of reviews was indispensable to the quality of the final collection. I also wish to thank Lauren Malone and Emma Ferguson for their ongoing and remarkably patient assistance in navigating the administrative and technical hurdles of a two-volume production.The theoretical anchor of this issue, the Kaplan Paradox, was sparked by Christopher Woffard’s 2022 Wired article, which provided the necessary vocabulary for the operational invisibility I have witnessed throughout my thirty years in the technology industry. However, a concept becomes a discipline only under empirical pressure. I am indebted to the authors of these papers for their intellectual courage in reframing their research to align with this “Hard Governance” architecture.Finally, I must acknowledge the peer reviewers. Although they remain anonymous, their contributions are evident on every page. In an era when managerial sedatives often pass for scholarship, these reviewers served as the primary auditors of the logic in this issue. Their rigorous, often necessarily uncomfortable critiques ensured that our Control Room was built on a diagnostic foundation, examining the institutional duty of care in an era of automated scale. Their labor is the silent, generative heartbeat of academic discipline.