Every major technological innovation has brought the same question with it: which jobs will disappear? It happened with mechanical looms, with electricity, with computers and with the internet. Generative artificial intelligence makes the question more pressing because, for the first time, it directly affects activities regarded as intellectual: writing, analysing, programming, designing. Answering with precise forecasts would be dishonest, because no one knows for sure how the labour market will evolve. But you can understand the mechanisms at work and prepare accordingly.
Tasks are automated, not jobs
A job is a bundle of different activities. An administrative clerk records invoices, answers suppliers, resolves discrepancies, talks to the accountant and organises deadlines. AI can speed up some of these activities, such as extracting data from documents or drafting routine emails, but hardly all of them.
The most useful way to reason is therefore to break your work down into tasks and ask, for each one, how exposed it is. The typical result is not "my job disappears" but "my job changes its composition": less time on some activities, more on others.
Which activities are most exposed
The activities where generative AI is already very capable share some common features:
- they consist mainly of producing or transforming text, code or images;
- they follow recurring, well-documented patterns;
- their result can be checked quickly;
- they require neither physical presence nor personal relationships of trust.
Examples: first drafts of standard documents, working translations, summaries, repetitive code, answers to frequently asked questions, product descriptions.
Which activities remain more human
Other activities are, at least for now, much less exposed:
- physical work in changing environments, such as maintaining installations or home care;
- relationships of trust: negotiating, advising, handling a conflict, supporting someone through a difficult moment;
- responsibility: signing off a project, deciding under uncertainty, answering for the consequences;
- judgement of context: understanding what a client really needs when they cannot explain it themselves;
- defining problems, before even solving them.
| Type of activity | Exposure to AI | Example |
|---|---|---|
| Producing standard texts | High | Reminder letters, minutes |
| Analysing structured data | Medium-high | Monthly sales reports |
| Client relationships | Low | Commercial negotiation |
| Skilled manual work | Low | Installing systems |
| Decisions with responsibility | Low | Approving a project |
The risk for newcomers
A less discussed effect concerns those entering the workforce. Many professions have always trained young people through simple, repetitive tasks: the first draft, document searches, basic code. If AI performs these tasks, opportunities for apprenticeship shrink. Forward-looking organisations will need to rethink internal training, giving young people the critical review of AI output and pairing them with experienced colleagues, so they can develop the judgement that no tool provides.
Skills worth developing
Using the tools, critically
You do not need to become a programmer. You need to be able to formulate clear requests, recognise AI's typical errors and know when not to use it. It is a practical skill acquired through guided use.
Domain expertise
Paradoxically, the more capable AI becomes, the more it matters to know your field well. Only someone who knows the subject notices that a summarised contract has lost a decisive clause or that an analysis has confused correlation with causation.
Communication and relationships
Explaining, listening, persuading, collaborating. These skills gain value precisely because producing standard content is becoming cheaper.
The ability to learn
Tools change quickly. More than any single technical skill, what counts is the habit of keeping up to date, experimenting and adapting the way you work.
What organisations can do
- Map tasks instead of reasoning in terms of whole roles.
- Involve people in choosing what to automate: those who do a job know where time is lost.
- Invest in training, not least because the AI Act requires those using AI systems to ensure an adequate level of AI literacy among staff.
- Redistribute the time freed up towards higher-value activities, rather than seeing it only as a cost cut.
- Protect the career paths of younger staff.
A realistic approach
Neither unbridled enthusiasm nor doom-mongering helps. Some activities will shrink, others will emerge, many will change shape. The difference, for people and for businesses, will be made by the ability to use these tools knowingly, keeping at the centre the skills that make human work hard to replace.
In short
- AI mainly automates specific tasks, not entire jobs.
- The most exposed activities are standard, verifiable text and code production.
- Relationships, responsibility, physical work and contextual judgement remain more human.
- It pays to develop critical use of tools, domain expertise and the ability to learn.
- Organisations must train people and protect younger workers' career paths.
