Will AI take my job?

What the data says

The impetus for this article was a conversation with friends one Saturday evening. Shortly afterwards I came across a podcast in which Marina Mogilko talks to Erik Brynjolfsson, an economist at Stanford University who has been studying the impact of technology on work for thirty years.

Brynjolfsson has published a study with the telling title “Canaries in the Coal Mine”. The study shows that in the occupations most exposed to AI, employment among people up to the age of 25 is well over ten per cent lower than the path of their less exposed peers would suggest. In the conversation the figure was 16 per cent, while the August 2026 update of the study already puts it at around nineteen per cent. In the firing line are, above all, younger programmers, call centre staff, some marketing roles and paralegals. Source: Brynjolfsson, E., Chandar, B., & Chen, R. (2025, revised 2026). Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence. Stanford Digital Economy Lab.

The same data, however, shows the other half of the picture. In the occupations least exposed to AI, employment is growing. Among experienced workers it is growing. And the most interesting thing: people who use AI to strengthen their own work, rather than merely having it do the work for them, are doing markedly better than the rest.

There is one more story in that conversation worth knowing. In 2017 Geoffrey Hinton, one of the fathers of deep learning, declared that we should stop training radiologists, because AI reads medical images better than humans. Today, as Brynjolfsson says, there are more radiologists than ever and there is still a shortage of them. Reading images is one of 26 tasks in that profession, and when a scan becomes cheaper, people simply have more scans done.

Brynjolfsson calls this the airline ticket principle: the jet engine lowered the price of a flight, and in response we fly so much that we spend more on flying than we used to. Cheaper knowledge work may therefore mean more work for the people who do it.

My personal point of view

The Stanford Digital Economy Lab data I cite above describes the whole economy, but the question “will AI take my job” always concerns a specific person. My personal view is this: whether or not you lose your job depends on how you build value using AI. From that point of view I distinguish four occupational groups.

Group one: people whose work does not pass through a computer. A plumber, a farmer, a building site crew. AI does not enter their working tools directly, so today it affects them at most indirectly. Brynjolfsson confirms this: physical work remains the hardest for machines, although in his view this window too will slowly close.

Group two: people who work at a computer but build no value with AI. They ask an AI chat about the weather and where to buy a “good drill”, and that is where their contact with this technology ends. If at the same time their work consists of repetitive activities in IT tools, such as handling tickets, simple programming, graphics, copying data between systems and producing content without producing a strategy, in my view they are the most at risk. These are precisely the activities AI agents are taking over fastest.

Group three: people who build value with AI for themselves. They raise the efficiency of their own work: they analyse faster, write faster, check faster. They have jumped onto a train that is just pulling out. What drives them is plain curiosity, and it is curiosity that decides who really learns these tools. These people will keep their position, because they will naturally replace those in group two.

Group four: people who build value with AI for others. They create products, reorganise processes and raise the efficiency of the organisation they work in or of their client's organisation. Brynjolfsson describes their future vividly: each of them will be like the head of a fleet of AI agents. The agents do the work, and the human is responsible for two things that cannot be done without a human: defining the right tasks and assessing whether the result is what was intended. This group will build the highest value and, in my view, will grow the most.

The three parts of every project

Take any project. It may be building a complex application, but equally a small thing needed in one's own work: combining two Excel spreadsheets and drawing conclusions, generating a third sheet containing data aggregated from the previous two, or even an ordinary recipe we want to refine.

Every such process, which above we called a “project”, consists of three parts: defining the task, carrying it out and assessing the result. Defining the task often involves working out a certain strategy, smaller or larger, but always a strategy. This is work that goes beyond a single instruction typed into an AI chat: it usually includes research based on the available data, from one's own resources and IT systems or from the internet, from product specifications, through competitor research, to what people write on online forums. Even this stage, with conscious use of AI, runs faster and reaches deeper, and at the same time gives the whole its context: thanks to it, the next steps go much more smoothly, because AI knows what it has to carry out in stage two. Then comes the execution, and finally the assessment of the result. In larger projects these activities loop into multiple iterations: after the result is assessed, the assumptions are redefined and the process starts again from the beginning, until the emerging product meets all the criteria.

Throughout history, humans performed all three activities. Today AI agents are taking over the middle one and performing it almost on their own. The two at either end remain, and they are the ones becoming valuable: the ability to name the problem before anyone starts solving it, and the ability to look at the result and ask whether it really is what we needed.

How this currently works for me

Take, for instance, the goidea.net website you are on right now. I built it in three months, together with the Knowledge section in six languages, working with AI very intensively, often on several threads in parallel. Today I develop it regularly, adding further articles. Each of them requires work that goes beyond the writing itself, in which AI helps: for example research on the market or on some other question, which brings context and only then makes it possible to build the content. Interestingly, coding the pages themselves, and even their design, was a small part of that work. Most of the time went on strategy, the offer, naming clients' problems and the content, together with checking it repeatedly and giving it direction, plus testing and further corrections after confronting it with real users.

A year earlier I would not have taken on such a project at all, because it would simply have been beyond me: I would not have managed to do my normal work and build such an extensive website at the same time. My previous site was a plain business card, because that was all I had time for. I built it over two weeks of afternoons, far faster than the initial version of the current site, but the quantity and quality of content on the two sites are not comparable: the difference is several-fold. AI has allowed me to completely change the way I communicate and has moved my work up a level, from doing to defining and assessing, and there is more work along the way, because more things have become possible at all. The airline ticket principle has been fully borne out in my case.

Where to start

If you feel you are in group two today, I have good news: it is still possible to board this train, and the ticket costs one afternoon. You need to board now, though, because it has already set off.

Brynjolfsson gives a method I recommend, because I have seen it work.

Moving from group two to group three is a matter of weeks, and moving from group three to group four begins the moment you first use your new tool to improve someone else's work.

To finish, I will repeat: this is my personal point of view and I will gladly set it against other opinions. Brynjolfsson ends the conversation with a sentence that also captures my own state of mind well: anyone who looks at AI without feeling both excitement and concern at once sees at most half the picture.

If you would like to talk about what your company looks like from this point of view and what could be built in it with AI, write to me or book a meeting.