"AI is going to destroy jobs."
We hear it so often by now that it almost sounds like an established fact. As if the future has already been written and we are merely waiting for it to play out. Entire professions will disappear. The discussion is barely about whether it will happen anymore, but mostly about when.
That is why it is worth looking back at a profession once predicted, with great confidence, to be the first to go: the radiologist.
A few years ago, when ChatGPT broke through, that story went viral. The reasoning was simple. Computers were learning to analyse medical scans and getting steadily better at it. In some tests they even outperformed human experts. An algorithm does not grow tired, does not get distracted and does not have a bad day. It does not miss a subtle anomaly because it has been staring at images for ten hours straight.
The conclusion seemed obvious. If a machine can read scans better than a human, why would we still train radiologists?
But then the thing happened that so often happens with predictions about technology: reality did not stick to the script.
Today, almost every modern scanner indeed contains some form of AI that helps doctors detect anomalies faster. At the same time, demand for radiologists is greater than ever. Both observations are true at once. AI has not replaced the radiologist. It has become a tool within the radiologist's work. As a result, specialists can handle more examinations, work faster and concentrate more fully on the complex cases where their expertise makes the most difference.
The mistake was not in the technology. That worked just fine. The mistake was in the assumption that the future develops in a straight line.
We do this constantly. We look at what is happening today and extend that line into tomorrow.
- AI reads scans, so radiologists disappear.
- AI writes code, so programmers disappear.
- AI generates text, so marketers disappear.
On paper that sounds logical. We just forget all the forces that bend such a straight line.
- When a tool becomes cheaper and more accessible, it usually gets used more, not less.
- When work becomes more efficient, demand for that work often grows.
- And when a skill becomes available to everyone, the bar for what counts as valuable tends to rise.
We have seen the same thing before. When spreadsheets arrived, many predicted accountants would become redundant. The opposite happened. Because calculations could suddenly be done quickly and cheaply, companies started asking far more financial questions. The need for financial expertise grew.
The content of the job changed too. The accountant spends less time calculating and more time analysing and advising. The radiologist spends less time on routine checks and more on complex diagnoses. What the predictions said would vanish often turns out, in reality, to be rearranged.
Jensen Huang, CEO of Nvidia and one of the principal architects of the current AI wave, put it well at Adobe Summit 2026. In his view, we often confuse the tasks of a job with the purpose of a job. AI is excellent at automating tasks, but far less good at taking over the underlying purpose a profession exists for.
The task of a radiologist is to analyse a scan. The purpose of a radiologist is to help diagnose and treat a patient. When the first task is automated, the second purpose does not disappear. On the contrary.
Perhaps that points to a more useful way of looking at the impact of AI. Not every job consists of a collection of separate tasks. Some professions are deeply anchored in a broader responsibility, a human relationship or a concrete outcome.
The jobs most vulnerable to automation are often the jobs whose entire value coincides with the execution of tasks. The jobs that endure are usually tied to a purpose that reaches beyond those tasks alone.
Maybe that is also why the question "Will AI replace my job?" is not particularly useful. No one can answer it with certainty today. And whoever claims they can usually overestimates their powers of prediction.
A more interesting question is: "How does my work change when this technology becomes available everywhere?"
That question starts from two assumptions that, historically, often prove correct.
- The first is that the technology will indeed become established.
- The second is that the work usually does not disappear entirely but takes a different form.
The attention then gradually shifts to something more useful: not defending the current way of working, but discovering the next version of it.
The same principle applies not only to professions, but also to companies and software products. When a new technology appears, the relevant question is not which features it can copy. The relevant question is which purpose the product helps achieve.
Software that essentially consists of screens, forms and reports, and of actions users must perform over and over, runs a risk. These are digital versions of jobs that consist mainly of tasks. AI is getting better and better at exactly that kind of activity.
But software that helps people reach an important goal is in a stronger position. Customers rarely buy software for the screens. They buy software because they want to generate more revenue, retain customers, lower costs or make better decisions. That underlying purpose remains, even when the technology changes.
That AI will take over many tasks is barely in dispute. But saying today that AI will destroy all jobs is the same error of reasoning as claiming AI will destroy no jobs at all. In the latter case you bury your head in the sand. In the former you opt for the intellectually laziest solution: pessimistic hopelessness. In both cases a complex future is reduced to a simple straight line.
The reality is that no one knows how this story ends. I do not know. Jensen Huang does not know. The journalist who splashes the most spectacular prediction across the page does not know either.
Spectacular doom scenarios get more attention than nuanced analyses. But getting attention and being right are two different things.
More than a century ago, Charles Darwin already noted something that remains relevant today. It is not our task to predict the future. Our task is to adapt to it.
That is not easy, because we feel uncertainty, and that runs against human nature. But perhaps there really is nothing else to do than what Rainer Maria Rilke advised a century ago: to be patient with all that is still unresolved, and to learn to cherish the questions themselves instead of chasing answers we cannot yet find.
And whoever keeps living the questions will, one distant day, without noticing it, grow into the answer.
