Workers are hiding the productivity gains they achieve with AI because they fear proving that their jobs could be done by fewer people, according to workplace research and analysis of how employers introduce new technology.
The dilemma is straightforward. Telling a manager that a report completed in an hour instead of an afternoon may bring recognition, better work or a promotion. But it could also prompt the question of whether a team of three is still needed.
Keeping quiet does not remove the risk. A colleague may discover a similar shortcut, leaving the original employee concerned that their advantage could eventually be used against them.
Glean, a workplace AI company, surveyed 6,000 mainly computer-based workers in the United States, the UK and Australia. Among those who said the technology had helped them complete more work or produce better results, 36% said they had concealed how much it had helped.
A 2024 survey by Microsoft and LinkedIn found a similar concern. More than half of employees using AI at work feared that relying on it for important tasks would make them appear replaceable.
Why workers are hiding AI productivity gains
The anxiety echoes a problem identified more than a century ago in factories. Frederick Winslow Taylor wrote in 1911 about workers who deliberately limited their output because they feared that higher production would lead employers to reduce the amount paid for each item.
They could then be forced to make more simply to maintain their previous earnings. Taylor called the practice “soldiering” and sought to eliminate it, while acknowledging that he would have resisted producing more if he had been in the workers’ position.
The same tension can emerge when managers ask staff to demonstrate their most effective AI techniques. For employers, the exercise can spread useful knowledge across a team. For employees, it may feel like an invitation to explain how the organisation could operate with fewer people.
Productivity gains can also be absorbed by additional work rather than shared with employees. Time saved through AI may be redirected into new assignments, leaving workers no better off despite completing their existing duties more quickly.
One example described by the leader of a Norwegian research centre involved using AI to work through budgets and prepare meeting minutes. The time recovered could subsequently disappear during a later budget review. A colleague using a similar shortcut to finish work on time faced the same problem if the reward was simply another task.
Norway’s approach to AI at work
Norway’s government and eight employer and union organisations signed an agreement on October 2 covering the use of AI through 2029. It calls for employees to be involved when the technology is introduced, using existing rights, and says the benefits should extend to workers as well as businesses and society.
The approach draws on Norway’s history of involving employees in the design of workplace technology. In the early 1970s, computer scientist Kristen Nygaard worked with the country’s Iron and Metal Workers’ Union on the introduction of computers, helping to establish the idea of participatory design.
The agreement is a statement of intent rather than a guarantee of higher pay or a share of savings. Its influence will depend on how it changes decisions inside workplaces.
Norway’s union membership also gives the arrangement a foundation that is less common in the United States. In 2025, 52.6% of Norwegian employees belonged to a union, compared with 10% of American workers and 5.9% of private-sector workers in the US.
Employers elsewhere could adopt the principle of involving staff without having the same means of enforcing it. The central question is whether employees are given a meaningful say in what AI will mean for their jobs, or are asked for their views only after decisions have already been made.
One possible model would be a six-month trial in which staff were promised that they would not lose pay or jobs as a result of the experiment. Employees could help decide how time savings were used, while managers would discuss the results with them before increasing targets.
Workers might choose paid training or time to clear backlogs that had led to late finishes. A company might instead use the capacity to take on more business. The important issue would be agreeing how the gains were used before they were committed elsewhere.
Such assurances would need to be written down and supported by a process for challenging a broken commitment. They could not prevent a company from later deciding to reduce its workforce, but they might give employees a reason to share the methods they have developed.
For employers, the response to AI may therefore depend not only on how much more work the technology makes possible, but on what happens to the people who reveal that potential. The willingness of employees to share their most effective shortcuts is likely to be shaped by how the previous productivity gains were treated.
