When Algorithms Rule the Office, Who Still Gets to Judge?

A recruitment specialist opens her dashboard one morning and finds an AI-generated ranking of job candidates. The system offers a score, but no explanation of how the variables were weighted. Company policy treats the ranking as the default; overriding it requires special justification. And yet, if the hire goes wrong, she is the one who will be held accountable. She may be well paid, satisfied with her job, and free of obvious stress, but something essential has quietly shifted: her professional judgement no longer carries real weight in the decision she must answer for. This unsettling scenario sits at the heart of a new conceptual paper published in the journal AI & Society by Mario Alberto Salazar-Altamirano of the Autonomous University of Tamaulipas and CETYS Universidad in Mexico, and Rafael Ravina-Ripoll of the University of Cádiz in Spain. Their argument is that workplace well-being research, for all its sophistication, has been measuring the wrong thing in the age of algorithmic management.

The two researchers introduce a construct they call Epistemic Well-being at Work, or EWW, defined as the organisational condition under which individuals and collectives can exercise responsible professional judgement, effective epistemic agency, and recognised epistemic responsibility within structures of authority. In plain terms, EWW asks whether workers can still produce, evaluate, contest, and justify knowledge claims that actually influence consequential decisions, and whether they can answer for those decisions in terms of reasons rather than mere metric compliance. Unlike hedonic well-being, which tracks pleasure and satisfaction, or eudaimonic well-being, which tracks meaning, growth, and autonomy, EWW is not a psychological state at all. It is a structural property of the organisation, expressed through governance practices, process design, incentive systems, and criteria of legitimacy. A workplace can be emotionally positive and epistemically fragile at the same time.

The paper’s central critique targets how organisational research has traditionally positioned workplace conditions. Decades of scholarship on leadership, organisational climate, justice, human resource practices, and work design have treated these factors as antecedents, resources, or moderators that explain an individual’s affective, evaluative, or functioning outcomes. Within the influential Job Demands-Resources model, for example, well-being emerges from the balance between demands and resources, predicting engagement, burnout, and performance. The authors do not dispute the value of this tradition. Their point is narrower and more radical: these models rarely ask whether organisational arrangements are constitutive of whether a worker’s reasons can enter, challenge, and alter consequential decisions. The missing level of analysis, they argue, is not the organisation itself but the epistemic architecture of organisational authority, meaning the structures that determine whose knowledge counts and whose can be overruled.

What makes this gap urgent is the arrival of intelligent technologies whose outputs acquire practical standing in organisational life. The paper’s scope covers algorithmic management and performance analytics, predictive scoring and automated decision-making, professional decision-support systems, workplace surveillance and people analytics, and generative AI when its outputs are treated as knowledge claims or decision inputs. Crucially, the authors distinguish between the technologies themselves and what they call AI-mediated governance. A predictive model is a technical artefact; governance is the institutional arrangement through which organisations give that model’s outputs a role in structuring behaviour, allocating resources, evaluating performance, or authorising decisions. The same system, technically identical in two organisations, may support judgement in one where its output is advisory, contestable, and tied to identifiable human responsibility, and displace judgement in another where it becomes an unquestioned default reinforced by incentives and penalties.

Drawing on recent work by Sergeeva, Leonardi, and Faraj, the authors adopt the view that intelligent technologies do not merely supply information but generate knowledge claims grounded in computational inference, potentially establishing competing epistemic regimes within organisations. These regimes redefine what counts as valid evidence and who is authorised to challenge it. When a dashboard or an automated recommendation acquires a status of technical objectivity, it becomes difficult to contest, and authority becomes less visible yet no less consequential. Algorithmic governance, in this reading, does not simply coordinate tasks; it structures the architecture of legitimate knowledge. Professional deliberation can become subordinated to outputs whose internal logic remains partially inaccessible, and this subordination is often normalised in the name of efficiency without anyone examining its implications for the humans expected to live with the consequences.

The authors identify three structural components of EWW. The first is integrity of judgement: the capacity to formulate, sustain, and justify professional decisions on reasoned grounds, even under structural pressure, with dissent not incurring disproportionate organisational cost. The second is effective epistemic agency: thinking alone is insufficient, because organisational architecture must not systematically neutralise human judgement in favour of automated outputs. Agency is effective only when professional reasoning can interrogate knowledge claims and meaningfully influence the final decision. The third is recognised epistemic responsibility: actors must be able to answer for their decisions in terms of reasons, not solely in terms of metric compliance. None of these components is a psychological trait. They are structural properties, and their erosion can be gradual, silent, and invisible to conventional satisfaction surveys.

Indeed, one of the paper’s most striking claims is that the erosion of EWW may coexist with high levels of affective well-being. The warning signs are not stress or dissatisfaction but increasing epistemic dependence on automated recommendations, reduced professional disagreement due to the structural costs of contestation, uncritical internalisation of metrics where compliance replaces deliberation, and displacement of responsibility, with decisions attributed to systems rather than accountable actors. These dynamics are frequently framed as efficiency, alignment, or successful technological adaptation. From a structural perspective, however, they may signal the weakening of the epistemic infrastructure that sustains professional dignity and responsible judgement. An organisation may run credible programmes to reduce burnout while leaving intact a decision architecture in which system outputs are presumptively correct and human overrides are penalised.

To connect these ideas to broader questions of sustainability, the authors propose the Human Sustainability and Epistemic Well-being Framework, an integrative architecture linking four nodes: organisational algorithmic governance, human dignity at work, EWW, and sustainable organisational well-being. The framework advances a conditional logic rather than a validated causal chain. Algorithmic governance configures the decision environment; EWW determines whether responsible judgement remains structurally viable; human dignity depends on that viability, because dignity requires being recognised as a reasoning agent capable of deliberation and accountability; and sustainable organisational well-being becomes more plausible when these conditions align. The authors are careful to note that the framework does not claim technology is inherently detrimental, that human judgement is invariably superior, or that EWW alone secures every dimension of well-being. Human decisions can be biased and inconsistent, and intelligent systems may improve accuracy and reveal errors. The relevant question is not human versus machine but whether their interaction is governed through contestable knowledge claims and aligned responsibility.

The practical implications reach the level of decision architecture rather than another wellness intervention. The authors argue that consequential AI-mediated recommendations should remain open to reasoned review through explicit override, appeal, and escalation routes, and that a formal right to speak is insufficient unless an alternative judgement can actually affect the outcome without triggering disproportionate penalties. Metrics should function as evidence rather than substitutes for judgement, with performance systems recording justified deviations and evaluating the quality of reasoning rather than treating deviation itself as error. Traceability should be calibrated to role and consequence, giving workers actionable intelligibility about inputs, assumptions, uncertainty, and limits, rather than demanding complete technical transparency in every case. And where organisations require humans to sign off on AI-assisted decisions, they must provide the corresponding capacity to interrogate and revise those outputs, because accountability without epistemic authority creates a structurally incoherent arrangement in which responsibility is retained while judgement is displaced.

The paper is explicitly conceptual, and the authors acknowledge its limits: no empirical evidence yet shows that EWW predicts dignity, legitimacy, or sustainable well-being, and the construct will require qualitative research, scale development, and longitudinal designs to establish that it is distinct from autonomy, psychological safety, and justice. They also caution that EWW may be experienced unevenly across hierarchy, occupation, and employment status, and that cross-national comparison, including in emerging economies and Global South contexts, is essential to prevent the framework from assuming highly regulated settings as the default. Even so, the contribution lands at a moment when algorithmic systems increasingly participate in cross-border decision-making, performance evaluation, and resource allocation, often without explicit deliberation about their implications. The paper’s closing question deserves to travel far beyond organisational theory: when we evaluate AI-mediated work, we should ask not only how efficiently decisions are made, but whether those expected to answer for them retain the authority and conditions to judge.

Subject of Research: Epistemic well-being at work under algorithmic governance and AI-mediated decision-making

Article Title: Beyond efficiency: epistemic well-being at work under algorithmic governance

Article References: Salazar-Altamirano, M. A., & Ravina-Ripoll, R. (2026). Beyond efficiency: epistemic well-being at work under algorithmic governance. AI & SOCIETY. https://doi.org/10.1007/s00146-026-03351-9

Image Credits: AI Generated

DOI: 10.1007/s00146-026-03351-9

Keywords: epistemic well-being, algorithmic governance, artificial intelligence, workplace well-being, epistemic agency, professional judgement, algorithmic management, human dignity, human sustainability, automated decision-making, organisational theory, AI & Society