Most AI programmes are announced before their human consequences are understood. A budget is approved, a platform is selected and a pilot is scheduled. Yet the organisation may still have no settled answer to the question employees will ask first: what will this mean for us? Efficiency, in the abstract, is not an objective. It is a direction of travel with no stated destination.

That gap is rarely deliberate, but it is not harmless. Leaders often set out the business case, the technology and the timetable, then delay the conversation about roles, workload and employment because the answers are still taking shape. Employees do not read that silence as caution. In the absence of a credible account of the human impact, they write their own version: this is about headcount.

The consequence is more than a communications problem at launch. It becomes an ongoing management problem, with consequences for adoption, wellbeing and retention. Trust depends on whether managers keep explaining what is known, what remains undecided and how people will be supported as the work changes.

Pew Research Center's September 2026 global research places that workplace anxiety in a wider public context. Across the countries surveyed, more people expect AI to cause job losses than to create jobs. That does not prove a particular rollout will reduce employment, but it changes the burden on leaders. Employees encounter AI against a widely shared expectation of displacement. Training can explain how a system works. It cannot explain what the organisation intends to do with the capacity it releases.

The exposure paradox

A CESifo working paper issued in May 2026, using Gallup’s workforce panel and some 82,945 pooled observations of American workers between 2023 and early 2026, found that frequent AI users reported greater fear of displacement than less-frequent users. In 2024, 8.6% of frequent users thought job elimination very likely within five years, against roughly 2% of everyone else.

The finding is counterintuitive only if exposure is assumed to bring reassurance. It may instead bring information. An employee who watches a system complete in ninety seconds a task that once took an afternoon has learned something concrete about that task's future. The same demonstration can produce enthusiasm for the tool and anxiety about the role.

Between those reactions sits an unanswered management question: the technology's capability is demonstrated daily, while the employee's future is described, if at all, in the language of strategy decks. Communication cannot promise certainty where none exists — but it can keep uncertainty from curdling into abandonment, simply by keeping the human consequences discussable.

The better the transformation, the sharper the fear

A Boston Consulting Group (BGC) 2025 survey of 10,635 leaders, managers and frontline employees across eleven countries found greater job-loss fear in firms redesigning whole workflows than in those deploying stand-alone tools: 46% against 34%. The organisations making the transformation most visible were also making its employment implications harder to ignore.

These were not simply negligent employers. The workflow-redesign group reported more training, stronger leadership support and better measurement of AI's financial contribution. Yet fear was higher. Good implementation can still create anxiety when the future operating model becomes clear faster than people's place within it.

This is the central failure: leaders explain the destination for the process, not the journey for the people. The question shifts from how do I use this tool to where do I fit once it is built — and a credible answer, whether reskilling, redeployment or an honest "we don't know yet," must never treat that question as illegitimate.

What training cannot do

There are two anxieties, not one. Capability anxiety asks: can I use this? Security anxiety asks: will the organisation still need me? Training answers the first with some reliability. It has little purchase on the second unless it is connected to clear decisions about roles, judgement, workload and development.

Effective training can even sharpen security anxiety, simply by proving how capable the technology is. A demonstration of the machine's power needs to be matched by a conversation about the person's future. Without that conversation, enablement can feel like preparation for one's own replacement.

BCG also found gaps in practical enablement: only 36% believed they had been adequately trained, roughly a quarter of frontline employees reported sufficient leadership guidance, and 37% said their employer had not provided the right tools. Weak enablement is an adoption problem; weak human-impact communication is a trust problem — and the two require different fixes.

A lesson from three closures

I saw this dynamic firsthand while managing large-scale operational closures in Japan, Indonesia and Hong Kong. Employees — myself included — were kept on because our operational knowledge was essential to a safe wind-down, even though we knew our jobs would disappear once the work was finished. Our focus naturally shifted from corporate goals to CVs, recruiters, family conversations and managing the transition ahead. This was not disloyalty; it was a rational response to changed incentives. Leaders could not alter that reality, but they did have a duty to communicate honestly and support their people through it.

The parallel to AI transformation is imperfect — closures have a defined end date, where AI initiatives carry a more lingering ambiguity — but the underlying mechanism is the same: companies need employees' active cooperation at precisely the moment those employees are wondering whether helping the change along threatens their own survival. Seen this way, self-protection is not defiance. Research shows AI anxiety drives workplace withdrawal mainly through emotional exhaustion, a pathway that supportive leadership can substantially ease. The managerial task, then, is not to manufacture optimism or silence concern. It is to make consequences discussable, spot strain early, and step in before self-protection turns into full disengagement.

Managers are continuous infrastructure

The encouraging finding is that anxiety is shaped as much by the organisational environment as by the technology itself — which gives managers a continuing role. They are present when new use cases appear, when responsibilities shift, when an efficiency gain quietly becomes a staffing question. A single launch event cannot carry that weight.

In the Gallup data, employees who most strongly agreed that their organisation cared about their wellbeing were around 6 to 6.8 percentage points less likely to report displacement risk. In within-person analysis, high organisational support reduced the additional fear carried by frequent AI users by as much as nine percentage points. These are associations, not proof of causality, but they point in a consistent direction: sincere care changes how technological change is interpreted.

Sentiment also moved with leadership support. Among frontline employees without clear support, 15% felt very positive about AI's effect on their enjoyment of work; with clear support, 55% did. Managers are not launch-day messengers. They are the continuing layer through which employees decide whether AI is an investment in them or a preparation for their replacement.

Three questions, repeatedly

A credible transformation conversation must answer three questions repeatedly, not once.

• What is happening to the company? Why AI is being introduced, what problem it is meant to solve and how success will be judged.
• What is happening to the work? Which processes, decisions, workloads and responsibilities are changing as experience accumulates.
• What is happening to me? What the changes mean for this role, these skills, this career and this job, including what is known and what is not yet known.

The third question is the one employees care about most, and the one organisations most often avoid. It cannot be answered fully at the start, because the answer will keep evolving. That is precisely why communication must continue throughout the rollout. Silence does not preserve optionality; it hands the explanation over to rumour.

Four practices follow:

• direct support towards the heaviest users, since exposure may increase rather than reduce concern;
• redesign workflows with employees, so participation provides both knowledge and agency;
• connect coaching to real decisions about where judgement stays human and how released capacity will be used;
• and review anxiety and wellbeing alongside adoption, because licence activation says nothing about whether trust is keeping pace.

The organisations that handle this decade well will not be those that eliminate anxiety. Some uncertainty is unavoidable when technology changes faster than operating models. They will be the organisations that refuse to leave employees alone with that uncertainty. Trust is built when managers keep the human impact visible, communicate honestly as decisions evolve, and show through repeated action that wellbeing is part of the implementation, not an afterthought to it.

References

• Christos A. Makridis, “The Sum of All (Workplace) Fears: How Managers Mediate the Fear of AI Job Displacement”, CESifo Working Paper No. 12678, May 2026.
• Christos Makridis, “Using AI More Does Not Reassure Workers, Managers Do”, Gallup, 9 September 2026.
• Vinciane Beauchene, Sylvain Duranton, Nipun Kalra and David Martin, “AI at Work: Momentum Builds, but Gaps Remain”, Boston Consulting Group, 26 June 2025.
• Chengcheng Sha, Lin Wang and Xiaolei Pan, “Does artificial intelligence (AI) anxiety increase employees’ deviant behavior toward the organization?” Journal of Psychology in Africa, 35(2), 207–214, 2025.
• Pew Research Center, “Globally, more people expect AI to cause job loss than growth”, 17 September 2026.

All figures cited are observational associations, not causal effects. The Hong Kong, Japan and Indonesia material is a personal management lens, used to interpret the research rather than as independent evidence.