56% against 21%: who is actually learning to use AI at work
What happened
Researchers at the University of Konstanz in Germany have published the third wave of a long-running survey on how people actually use artificial intelligence at work. The results, released on 18 September 2026, are a useful corrective to the general noise on the subject. The survey questioned 1,105 employees in May 2026. It is part of a study that has run since March 2024, when 2,019 people were surveyed, followed by 1,024 in May 2025. The headline number is smaller than most people would guess. The share of employees using AI in their work rose from 35% to 38% over the year. That is growth, but it is slow growth. The second finding is about who decided. Only 55% of AI users said the tool they use most often was officially introduced by their employer. For the other 45%, the employee brought it in themselves. The third finding is the gap. Office and knowledge workers: 49% use AI. Production and manual workers: 25%. People with higher education: 56%. People with lower qualifications: 21%. In small organisations, just 11% of staff had received any AI training, and only 10% worked somewhere with binding rules on AI use. Professor Florian Kunze, who heads the Future of Work Lab at Konstanz, summarised it: "In many organisations, the major AI revolution has not unfolded as a planned transformation process so far." The research team also includes doctoral researchers Carolina Opitz and Elena Gerdiken.
Why this matters
This is an education story as much as a technology one, and the education part is the gap between 56% and 21%. If the people who already have the most qualifications are also the ones learning the newest tools at work, then a skills gap that already exists gets wider. It is not that anyone was excluded on purpose. It is that when nobody organises the training, the people who pick it up are the ones with the time, the confidence and the kind of job where experimenting costs nothing. That is what the 45% figure really means. When almost half of AI use is brought in by employees themselves, the organisation is not spreading a skill — it is watching a skill spread unevenly. The figures on small organisations follow the same logic. One in ten workers in a small firm has had training. One in ten works under any rules. That leaves most people improvising. There is something reassuring in the survey too. While 40% of employees worried about what AI might do to the labour market generally, only 17% feared losing their own job. People are more worried in the abstract than about themselves.
What the evidence actually says
The most important caveat first: this is a survey of employees in Germany. The UK labour market is not identical, and the percentages should not be quoted as British figures. That said, the structure of the finding is likely to travel. The pattern — knowledge workers ahead of manual workers, graduates ahead of non-graduates, large employers ahead of small ones — is not a German peculiarity. It is what you would expect anywhere that adoption is left to individuals. The survey's real strength is that it has run three times with comparable questions. A single snapshot tells you a number; three tell you a direction. And the direction here is a modest rise with a gap that is not closing. It is also worth being careful about what 'using AI at work' means. The survey measures whether people use it, not whether it helps them. Someone using a chatbot to rewrite emails and someone using AI to analyse data both count as users. Adoption is not the same as benefit. The finding about informal adoption deserves one honest note of caution that the researchers themselves point at. When staff bring in their own tools, they may be putting work information into services their employer has not checked. With only 10% of small organisations having binding rules, a lot of that is happening without anyone deciding it should.
Practical advice
For anyone at work: Ask what your employer's rules are before using an AI tool for work. If the answer is that there are none, that is worth knowing too — it means the responsibility currently sits with you. Do not put confidential information into a tool nobody has approved. Customer details, patient information, unpublished figures, anything covered by a contract. This is the single biggest practical risk in the whole survey. If you have taught yourself something useful, offer to show a colleague. In an organisation where 45% of adoption is informal, one person sharing what they know is how the gap starts to close. For employers, particularly small ones: Write something down, even one page. What tools are allowed, what must never be pasted into them, who to ask. Ten per cent of small organisations have this. It costs an afternoon. Offer training to the people least likely to ask for it. The gap in this survey is not about aptitude. It is about who gets the chance. Do not assume adoption means benefit. Ask staff what is actually saving them time. For anyone worried about their own job: the survey found more people worried about the labour market in general than about themselves. If you want to feel less worried, learning to use the tools is a more useful response than avoiding them.
What to know
A survey of 1,105 employees in Germany, published on 18 September 2026, found AI use at work rose only from 35% to 38% in a year. Nearly half of the tools people use were brought in by employees rather than employers. Use runs at 56% among the highly educated and 21% among those with lower qualifications, and at 49% for office workers against 25% for production and manual workers. In small organisations, only 11% of staff had any AI training and only 10% had binding rules to follow. The figures are German, but the shape of the problem is not. Where training is not organised, the people who already have the most qualifications are the ones who gain the new skill — and a workplace with no rules is a workplace where staff are guessing about what is safe to share. Sources: University of Konstanz via EurekAlert!, 'The use of AI at work: slow, informal and unequal', 18 September 2026, https://www.eurekalert.org/news-releases/1144563; University of Konstanz, 'AI in the workplace - growing usage, persistent inequality', https://www.uni-konstanz.de/en/university/news-and-media/current-announcements/news-in-detail/ai-in-the-workplace-growing-usage-persistent-inequality/; Arntz M and colleagues, 'Low Barriers, High Stakes: Formal and Informal Diffusion of AI in the Workplace', https://doi.org/10.2139/ssrn.6120067 This article is for general information only.
Related articles
Mondays, Wednesdays and Fridays: the hub that opened at a football ground
On 12 October 2026, the Department for Work and Pensions announced that a new Youth Hub had opened in Exeter. The hub is for people aged 16 to 24. It is at St…
A 'veneer of progress'? The youth justice chief answers her critics
On 12 October 2026, the chief executive of the Youth Justice Board for England and Wales published an article defending the approach that now shapes how the sys…
Three years of missing accounts, and two trustees on the register
On 12 October 2026, the Charity Commission announced that it has opened a statutory inquiry into an education charity called Crossfields Institute. The charity…