A research team led by the University of Bath's School of Management has published findings in the Academy of Management Review that should interest anyone who reports to, or is responsible for developing, managers in a British workplace. The study identifies a quiet capability risk: when managers routinely offload their thinking to generative AI tools, they gradually lose the practical judgement that their experience was supposed to build. The researchers call this process epistemic de-skilling, and their argument is that it is already under way.

What the Bath researchers actually found

The study centres on what the authors call managerial phronesis: the practical wisdom a manager develops through lived experience, reflection, and direct human interaction. It is the quality that lets a capable team leader read a difficult situation, weigh competing pressures, and make a call that no policy template can make for them. The paper argues that generative AI, used as an answer-dispenser rather than a thinking aid, has begun to hollow out that capacity.

Professor Dirk Lindebaum of the University of Bath's School of Management describes the mechanism directly:

"As managers increasingly outsource their thinking to Gen-AI for idea generation, or when a practical problem arises at work, they may rely less on their own judgment."

Lindebaum identifies the compounding effect as the central concern: "Over time, this could reduce their ability to learn from experience, think critically, and to anticipate what kinds of actions are needed now to meet future goals."

The paper is not a blanket critique of AI in management. It draws a deliberate contrast between epistemic de-skilling and its counterpart, epistemic upskilling, where AI serves as a reflection tool that sharpens human judgement rather than substituting for it. The researchers identify accountability as the pivotal design variable: upskilling is most likely when managers know they will be held responsible for decisions and expected to articulate their reasoning.

For employees: what an AI-dependent manager looks like from below

If your manager has quietly shifted their decision-making into a chat window, you are likely to notice the effects before anyone in a senior HR role does. The signals are recognisable and specific rather than dramatic.

• Feedback on your work feels generic, as though it was drafted with no reference to your particular role or context, because it probably was not.
• Decisions arrive fully formed but without visible reasoning, and the logic shifts when a follow-up question exposes a gap.
• Judgement calls that used to be handled quickly, a leave request, a client escalation, a workload conflict, now get delayed while your manager consults something they previously knew on the spot.

Lindebaum's team frames the downstream risk in terms that are directly relevant to employees: rather than developing a nuanced understanding of individual team members, customers, or organisational challenges, a de-skilled manager may lean on AI-generated answers that lack the contextual and moral weight that complex decisions require. The people who absorb the consequences of those decisions are, in most cases, the people below them.

Your options for managing this are narrower than the problem, but they are not trivial. When a decision affects your workload, your role, or your pay, ask to understand the reasoning rather than simply accepting the outcome. In one-to-one meetings, invite your manager to talk through how they weighed competing considerations. If a performance conversation feels boilerplate, request specific examples tied to your actual work rather than generic observations. Keep your own contemporaneous record of what was decided and the rationale given: that record matters if the stated reasons shift later. Under the Equality Act 2010, decisions that affect pay, promotion, or performance ratings must be capable of withstanding scrutiny on grounds including protected characteristics; a contemporaneous record of reasoning supports that scrutiny from both sides.

For HR: designing accountability before de-skilling appears in performance data

The uncomfortable aspect of this research for people professionals is that epistemic de-skilling does not announce itself in a quarterly dashboard. It surfaces as slower incident response, thinner coaching conversations, a rise in escalations that used to be resolved at team-leader level, and eventually in retention data when high-performing employees conclude their manager is not adding sufficient value. By the time the pattern is legible, it has typically been building for months.

The Chartered Institute of Personnel and Development (CIPD), whose annual surveys consistently identify manager capability as a primary driver of employee engagement and retention in UK organisations, has highlighted the risk that AI tool adoption in management outpaces investment in the human skills those tools are supposed to augment. The Bath paper's framing gives that concern a specific mechanism and a name.

Lindebaum's core recommendation is a design question rather than a training one: organisations need to carefully design roles, responsibilities and workflows to ensure employees continue developing the human skills that AI cannot replicate. Three areas where that design work translates into concrete moves for UK people teams:

1. Build reasoning into the record

Where managers make decisions that affect individuals, hiring, promotion, disciplinary action, performance ratings, they should be required to document the reasoning, not only the outcome. This is already sound defensive practice for employment tribunal purposes under the Employment Rights Act 1996, which provides a framework for assessing whether dismissal or adverse treatment was procedurally fair. The Advisory, Conciliation and Arbitration Service (ACAS) Code of Practice on Disciplinary and Grievance Procedures reinforces the same principle: written reasoning is not bureaucracy, it is evidence. It also happens to be the single most effective structural nudge toward epistemic upskilling: managers who know they will be asked to justify a decision tend to think more carefully before they make it.

2. Position AI as a challenger, not an oracle

In manager enablement programmes and internal AI-use guidance, framing matters. Lindebaum's recommendation is direct: rather than accepting AI outputs at face value, managers can use them to challenge assumptions, explore alternative scenarios, and test the reasoning behind their own decisions. He goes further on why the friction is productive: because AI systems often struggle to explain why they produce particular answers, the gaps in those explanations can encourage people to think more deeply about their choices and the consequences of their actions. Internal policies that treat AI output as a first draft to be interrogated will, over time, produce sharper managers than policies that treat it as a finished answer to be forwarded.

3. Redesign roles so judgement stays exercised

If most of the recurring decisions a manager used to make are now automatable, the role has quietly become an approval function. Approval functions atrophy quickly. Consider which parts of a supervisory role should be preserved as human-judgement work by deliberate design: coaching conversations, cross-team negotiations, ambiguous client situations, decisions with ethical trade-offs. Protecting those from workflow automation, even where automation is technically feasible, is where practical wisdom gets built and maintained. The CIPD's Good Work Index framework for manager quality offers a useful reference point for identifying which supervisory behaviours most directly affect employee wellbeing and performance outcomes in UK contexts.

The wider signal this study leaves open

The Bath paper identifies a direction of travel rather than a measured headcount of affected managers. The scale of actual generative AI adoption inside UK organisations, and the proportion of workers who currently report to a supervisor whose judgement has materially shifted as a result, is not quantified in the published findings. What Lindebaum's team has done is name a mechanism precisely: epistemic de-skilling as a structural consequence of how AI is deployed in management roles, not an individual failing. The counterpart, epistemic upskilling, is equally structural: it depends on accountability by design, not on individual willpower.

For UK organisations navigating obligations under the Employment Rights Act 1996 and the ACAS Code while also investing in AI-enabled management tools, the two agendas are not separate. The documentation requirements that make dismissal decisions defensible at tribunal are functionally the same accountability structures that the Bath researchers identify as the primary guard against epistemic de-skilling. HR teams that connect those two imperatives, rather than treating AI governance and employment-law compliance as separate workstreams, are better placed to catch capability drift before it becomes a people-data problem.

For employees, the immediate implication is that the manager you work under next year may be materially better or materially worse than the one you work under now, depending on design choices being made elsewhere in the organisation. The guardrails a CIPD-aligned people team puts in place in the current cycle, requiring reasoning to be recorded, framing AI as a challenger, preserving human-judgement roles by deliberate policy, are the ones that determine which way that drift goes.