Employees who reported working more closely with generative artificial intelligence were also more likely to see it as a professional opportunity and to report stronger engagement at work, according to doctoral research from Finland.
In many workplaces, the immediate change is happening inside jobs. Some employees are using AI to research, summarise, draft or solve routine problems, while colleagues doing similar work are not.
The resulting advantage is unlikely to come from using AI as often as possible. It is more likely to come from knowing which tasks it can improve, how to check its output and when human judgement should take over.
What the University of Vaasa research found
Zhe Zhu’s 2026 doctoral dissertation at the University of Vaasa examined generative AI in organisational decision-making and employees’ working lives.
One study used valid survey responses from 395 employed people in the United States whose workflows included generative AI. The researchers measured what they called AI collaboration, meaning that workers used AI to obtain knowledge, solve problems and support decisions.
Greater self-reported collaboration was associated with a stronger tendency to view AI as an opportunity for learning, growth and better performance. That opportunity appraisal was, in turn, associated with greater work engagement, including concentration, dedication and a sense that the job had meaning.
Seeing AI as a threat was associated with lower engagement. However, greater AI collaboration was not significantly associated with stronger threat perceptions in the main statistical model.
Job insecurity complicated the picture. Among workers who felt less secure about their future employment, the positive association between AI collaboration and opportunity perceptions was weaker, while the association with threat perceptions was stronger. This finding does not support the simple idea that fear reliably pushes employees to embrace AI. Pressure may make the same technology look less like a route to development and more like a warning about replacement.
A second survey in the dissertation covered 361 expatriate professionals who used generative AI in their work. More AI collaboration was associated with greater career adaptability, a research concept covering a person’s readiness to plan ahead, take control, explore possibilities and deal confidently with change.
AI collaboration did not have a statistically significant direct relationship with career sustainability, which the study measured through life satisfaction, career satisfaction and productivity. The statistical model instead found an indirect association through adaptability. In practical terms, merely using AI was not linked directly to a more sustainable career. The potential benefit appeared alongside workers’ capacity to adjust and learn.
AI can shorten part of the learning curve
Separate workplace evidence shows why effective use can matter. A study published in The Quarterly Journal of Economics examined the phased introduction of an AI assistant among 5,172 customer-support agents at a company selling business-process software.
The assistant monitored customer chats and suggested replies, but agents remained responsible for the conversation and could edit or ignore its advice. The researchers estimated that access to the system increased productivity by an average of 15%, measured as customer issues successfully resolved per hour.
The average concealed a large difference. Less experienced and lower-performing agents recorded the biggest gains, improving both speed and quality. Agents with two months of experience and AI assistance performed about as well as, or better than, unsupported agents with more than six months of experience. The most experienced and highest-performing agents gained little in overall productivity, with small improvements in speed and small declines in conversation quality.
This result challenges the idea that AI will always widen the gap between stronger and weaker workers. In this particular setting, it compressed part of the experience curve by making some practices used by better agents available to newer colleagues.
The researchers also found evidence consistent with learning. During occasional system outages, workers who had previously used the assistant continued to handle chats faster than before its introduction. The estimates were noisy because outages were rare, so this should not be treated as proof that every AI tool teaches transferable skills. It does suggest that a well-designed assistant can sometimes help people learn, rather than simply completing work for them.
Exposure does not mean that a whole job can be automated
The labour-market picture remains much broader than any single company or survey. The International Labour Organization’s 2025 global index estimated that one in four workers was employed in an occupation with some exposure to generative AI. The share was 34% in high-income countries and 11% in low-income countries.
Exposure is not a forecast of job losses. It means that an occupation contains tasks which generative AI may be able to perform or assist with. The ILO concluded that transformation was more likely than complete replacement because most occupations still include work requiring human input.
This distinction matters. An accountant, marketer or manager does not perform one indivisible activity. Each job combines routine and unusual tasks, communication, responsibility and judgement. AI may handle some of those elements well, perform others unreliably and be unsuitable for work involving confidential information or accountable decisions.
A job can therefore change significantly without its title disappearing. Employers may raise expected output, reorganise roles or reduce the time allowed for particular tasks. Workers may also spend less time producing first drafts and more time checking, interpreting and explaining results.
The real divide may be between managed and unmanaged adoption
Taken together, the evidence points to a more nuanced competitive divide than AI users against non-users. The customer-support study shows that access to a suitable assistant can help newer workers catch up. The Vaasa surveys indicate that perceived opportunity, adaptability and trust are associated with how workers respond. Neither finding suggests that buying employees an AI subscription is enough.
Companies still need to identify suitable tasks, protect sensitive data, train staff and establish who checks important outputs. They also need to avoid treating faster production as useful productivity when errors create extra work later.
There is a further lesson for managers. If insecurity weakens employees’ tendency to see AI as an opportunity, introducing the technology mainly through warnings about falling behind could be counterproductive. Training that gives workers practical experience, clear boundaries and responsibility for final decisions may be more useful than pressure alone. That is an inference from the research, not a tested management intervention.
The limitations are substantial. Both Vaasa employee studies used cross-sectional, self-reported surveys, so they cannot establish cause and effect. More adaptable and engaged people may be more willing to use AI in the first place. The customer-support findings are stronger workplace evidence, but they concern one tool, one company and one occupation.
The evidence therefore does not support the slogan that a colleague using AI will take somebody else’s job. It supports a narrower conclusion: when AI fits the task and people learn to use it critically, it can alter performance and shorten parts of the learning curve. The durable advantage is likely to belong to workers who combine the technology with subject knowledge, verification and sound judgement.
