Ai does not affect everyone equally: this is how the world of work becomes polarised

The adoption of AI in the workplace is changing the rules of the game, but not all professions are playing the same game. According to a study signed by Bouke Klein Teeselink, of King's College London, and Daniel Carey, of the AI Objectives Institute, analysing hundreds of millions of job advertisements in 39 countries before and after the public release of ChatGPT in November 2022, the impact of AI on job markets depends crucially on which tasks are automated.

While it is true that in the case of occupations highly exposed to AI, new job vacancies fall by an average of 6.1 per cent, it is also true that the effects vary depending on the type of tasks involved, and the political and social context and level of digital development in individual countries. Where employment protection is stricter, in fact, the drop in the labour supply is greater because companies where most of the tasks are performed are highly exposed to the use of artificial intelligence and hesitate to hire if they fear difficult and costly future redundancies; where the digital infrastructure is more mature, the contraction is smaller, suggesting that readiness to adopt and adapt to technology could help reduce unemployment.

The heart of Teeselink and Carey's investigation, then, is a point often overlooked in the debate: it is not just 'how much' AI can make of a job that matters, but 'which' pieces it replaces. When AI automates high-skill tasks, the remaining work gets easier: there is more supply of potentially suitable workers, and wages tend to fall. Conversely, if AI automates the routine and less-skilled parts of a job, the remaining tasks tend to be more specialised: fewer people can perform them, and thus wages rise. In this regard, the researchers cite the example of a human resources specialist whose administrative records are now handled by AI, leaving him free to concentrate on complex employee relations and decision-making. But when AI can perform more specialised and cognitively demanding tasks, wages decrease because the work no longer requires limited skills. This is the case, the researchers write, for roles such as junior software engineer.

For scholars, therefore, the same level of exposure to AI can produce opposite outcomes on wages and employment if artificial intelligence takes away different parts of the job. For the policy maker, this means that responses cannot be the same for all circumstances: one has to look at which segments of labour are at risk and calibrate training and protections accordingly. On the other hand, artificial intelligence has become an integral part of the world of work. The latest Gallup data, for example, tell of how AI adoption in the United States is advancing.

According to the US Institute's analysis, daily use of AI rose from 10% to 12% between the third and fourth quarters of 2025, while frequent use (i.e. at least a few times a week) reached 26%. Overall adoption, however, remains stuck at around 46%: the base is not broadening, but those who already use it are exploiting it more and more. This dynamic is particularly evident by level and role: managers and profiles with decision-making autonomy remain the most active.