I have worked as a digital entrepreneur in the tech world since 2007. In almost twenty years I have seen a lot of developments up close. Social media came of age, the smartphone changed our daily behaviour, e-commerce transformed entire markets and cloud software changed the way companies are built. Every new wave was first underestimated, then overestimated, and eventually simply became part of everyday life. I have always tried to understand early what such a development would really change, not only for my companies, but for people and for the society we live in.
AI feels different to me. Not because it is the next big technology, but because this technology moves directly onto the ground we always considered human. Writing, programming, analysing, designing, researching, advising and preparing decisions. More and more of that can already be done by an AI system, and those systems no longer wait for a single instruction either. They can carry out tasks independently, gather information, make choices within an assignment and handle longer processes. That means AI touches not just our productivity. It touches the economic value of human intelligence itself.
In early 2026 I read The Adolescence of Technology by Dario Amodei, CEO of Anthropic. That essay was an important reason for me to put my own thoughts on paper. Amodei does not write as an outsider trying to predict what AI could one day become. He stands in the middle of its development. He describes powerful AI as something that could eventually resemble a country of geniuses in a datacenter, millions of digital thinkers working at a high level, all at once. According to him this could become reality much sooner than most people think, although he rightly adds that nobody knows the exact timing.
What I take most seriously is the pace. In earlier technological revolutions, development ultimately remained dependent on people. People wrote the software, did the research, built teams, raised capital and set up companies. AI is now beginning to help develop AI itself. Once software helps write better software, and research helps accelerate research that in turn leads to stronger systems, the speed can keep climbing. That does not automatically mean every prediction will come true. It does mean that our usual way of thinking in years and generations can quickly become outdated.
His essay goes well beyond jobs, too. Amodei writes about cyberattacks that could become far more powerful, biological risks, authoritarian surveillance, autonomous weapons and the concentration of economic and political power. These are different problems, but they all arise from the same new fact. A small group of people may soon have access to an amount of digital intelligence that was previously conceivable only for enormous organisations or entire states. That calls for different rules and, above all, for leaders who decide in advance what they do and do not find acceptable.
Amodei warned earlier that AI could displace a large share of entry-level white-collar jobs within one to five years. That is a prediction, not an established fact. In March 2026 Anthropic itself published research that showed no systematic rise in unemployment yet in the occupations most exposed to AI. There were signs, though, that the inflow of young workers into some of those occupations was slowing. To me, precisely that nuance makes his warning more credible. We do not need to invent a jobs crisis before it exists, but it would be just as unwise to start thinking about the consequences only after the turning point.
Elon Musk has been talking about a similar direction for years. He predicted that AI would become smarter than any individual human and sketches a future in which robots and AI take over so much production that work could eventually become a choice. He uses the term universal high income for it and speaks of an economy of abundance. Musk has named deadlines before that did not come true, so I attach little value to an exact year. The timelines circulating on social media, with superhuman AI in 2026, fully automated digital work shortly after and hundreds of millions of humanoid robots in the years that follow, should not be treated as a schedule that is already fixed.
The direction behind such predictions, however, I do find relevant. Suppose this development takes not nine years but twenty. Even then we have to think about an economy in which fewer and fewer human hours are needed to produce more and more value. What happens to the first years of a career when precisely the work on which young people normally learn is largely automated. What happens to wages when digital labour becomes cheaper. And what happens to the distribution of wealth when the greatest economic value shifts from labour to the ownership of models, data, infrastructure, companies and other productive assets.
I am not writing this out of resistance to AI. I have built with technology my whole career and I use AI every day. The possibilities are enormous. Companies can build things with smaller teams that used to require whole departments. Research can be faster, education more personal, and medical knowledge far more accessible. For entrepreneurs, a period is beginning in which ideas can be executed at a speed and scale that was impossible only a few years ago. I feel that enthusiasm myself.
At the same time, we have reached a point where enthusiasm alone is no longer enough. That became very concrete for me when I saw NAZA, the documentary by Yuval Abraham and Rachel Szor produced by The Guardian. The film is based on the testimonies of 24 anonymous Israeli soldiers and intelligence officers. They describe systems that were used for surveillance, target selection and the estimation of expected civilian casualties in Gaza. Some of those systems work with AI. The term NAZA is used in the film for the number of civilians expected to die in an attack.
I found that hard to let go of. We are no longer talking about an algorithm that selects advertisements or helps a company work more efficiently. We are talking about technology that helps determine who comes into view as a potential target, which location is recommended and how many civilian casualties are weighed into the decision. In the film, one source even speaks of very high numbers of expected civilian casualties that could be accepted for an attack on a single target. Once AI selects and ranks people at scale in a process that can end with a missile on a home, it is no longer administrative support. The system is substantively steering the decision.
The Israeli army disputes the notion that AI independently decides who is attacked. According to the IDF, systems like Lavender are tools for analysts, targets must be assessed individually and the final decision remains with human commanders. That distinction is important. It just does not take away my concern. If a system can generate thousands of names in a short time and a human then mainly checks whether a recommendation meets minimum conditions, human control can very quickly turn into human confirmation. On paper, a human remains in the loop. In practice, the machine has already decided what that human looks at.
For me, that is where an ethical line lies. A human life must never be treated as a data point to be processed through a decision chain as efficiently as possible. Speed and scale are valuable in a business, but as soon as a decision touches someone's freedom, safety or life, care must weigh more heavily. That also means that some decisions are, as a matter of principle, not left entirely to machines, even when the technology seems to perform faster or statistically better.
That is precisely why I find the conflict between Anthropic and the Pentagon so interesting. Anthropic is not against military use of AI. The company has said explicitly that it supports many lawful applications for national security. It did draw two lines. Claude may not be used for mass domestic surveillance of Americans, nor for fully autonomous weapons that select and attack targets without meaningful human intervention. When the Pentagon demanded that those restrictions be dropped for all lawful applications, Anthropic refused. The conflict escalated to the point where the Pentagon designated Anthropic a supply-chain risk.
Whatever you think of Anthropic, there is an important principle in that for me. Ethical standards only mean something if you are willing to put something at stake for them. Otherwise they are conditions that disappear as soon as the customer is big enough. Lawful, moreover, is not automatically responsible. Legislation almost by definition lags behind technology. Companies building systems with ever more influence over work, information, safety and ultimately human lives will therefore have to set their own limits before a judge or a government does it for them.
That responsibility does not lie with AI companies alone. Leaders who buy and deploy AI must also know where their line is. It is too easy to shift responsibility onto a model, a vendor or an algorithm once an outcome goes wrong. A machine has no conscience and cannot bear moral responsibility. People can. Whoever decides to use a system remains responsible for what that system does within the space it is given.
That brings me to another subject that I believe gets too little attention, namely ownership. As intelligence becomes ever cheaper and more scalable, the value of the infrastructure around it rises. The models, chips, data centres, data, distribution and companies that own this capacity can attract an ever larger share of economic value. Amodei warns about this himself. He writes about a future in which AI companies and their owners could build up economic power on a scale we barely know today.
For entrepreneurs, that changes how I look at value creation. My generation first learned to scale with people, then with software and platforms. Now we can also scale intelligence. That offers enormous opportunities, but it also means ownership may become more important relative to selling time. Someone who only sells their hours stands differently in such an economy than someone who owns a company, technology, intellectual property, real estate, a brand or other assets that can keep producing value on their own.
That is also why I have started thinking differently about what we should pass on to our children. Educating them only for a good job feels too narrow for the world that is coming. They need to understand how technology works, how value is created, how you build something and why ownership matters. At least as important is that they learn to doubt a convincing answer, check sources and take responsibility for a decision themselves. As knowledge gets ever cheaper, the ability to handle it becomes all the more valuable.
The same goes for leaders. Ignoring AI is no longer a serious option. An organisation that does not understand this technology is going to fall behind. Automating everything because it is technically possible is not wise either. Good leadership will soon consist, to an important degree, of the choices you do not automate, the questions for which you make human judgement mandatory, and the limits you set in advance instead of afterwards, when something goes wrong.
I expect AI to become one of the greatest sources of productivity growth and scientific progress of our time. I also expect the same technology to be used for surveillance, manipulation, cyberattacks, weapons and the concentration of power. Both developments can exist side by side. That is why I have little use for the simple camp thinking between people who see AI only as an opportunity and people who are mainly afraid of what could go wrong. Those who build should see both sides.
After almost twenty years in technology, that is the most important lesson for me. Progress does not release us from responsibility. The more powerful our systems become, the clearer our own standards have to be. In the coming years we are going to make difficult choices about work, income, ownership, education, privacy, war and the role people themselves still play in decisions that used to be self-evidently human.
The real test, therefore, does not lie in whether we can build ever more intelligent machines. We are already doing that. The test is whether we, as entrepreneurs, leaders, governments and society, are mature enough to draw lines at the moment when speed, profit or power pull us the other way. AI can take over more and more of our work. Our conscience is not part of that.