I have a son
who's 4.
Here's his world.

A question that keeps me up at night

A few days ago, I was talking with my brother. We were talking about our kids, and thinking about the world they're going to live in. In the middle of that conversation, I realized something that gave me vertigo.

My son is four. When he's fourteen, it'll be 2036. When he's twenty-four and starting his adult life, it'll be 2046. I have no idea what the world will look like then. And not in the vague sense that no one ever knows what tomorrow holds. In the very concrete sense that I'm starting to doubt whether the reference points we teach our kids today will still mean anything when they're adults.

Will he learn to drive? Will he go to the doctor? Will he have a boss, a job, an office? Will his teachers be human?

I spent the last three weeks reading everything serious I could find on the subject. Engineers, philosophers, economists, researchers. I asked Claude to explore five different angles to dig into it with me. Here I'm sharing what stands out. No jargon, no needless complexity. Just what I understood, and why this reading is pushing me to rethink what I'm going to pass on to my son.

I could be wrong on many points. But I think we're underestimating what's coming.

What's already happening, today

Before heading into the future, we need to understand where we are. Because we're already somewhere other than where a lot of people think.

AI today — Claude, ChatGPT, Gemini, the others — these aren't gadgets for geeks anymore. One person in two worldwide uses it. Faster than anything we've known before. Faster than the smartphone, faster than the internet, faster than television.

And these AIs are already passing exams that 99% of humans would fail. They write code better than most developers. They answer medical questions better than a lot of general practitioners. They translate, they summarize, they analyze, they create images, voices, videos we can no longer tell apart from the real thing.

Let me be clear right away: they still get things wrong often. They make things up. They struggle with long tasks that require hours of focus. They can't yet really "work on their own" for a whole day like a human.

But here's the detail that changes everything: they're improving at a speed we've never seen. The researchers at METR (Model Evaluation & Threat Research) measure what an AI can do on its own before losing the thread. Today it's about one hour of human work. Two years ago it was a few minutes. That number doubles every four months.

If you project that pace over a few years, you understand why I'm asking myself questions.

My 5 detailed bets below · revised every year

Beyond AI itself, the same moves show up on the geopolitical side. In March 2026, Beijing published its 15th five-year plan with exactly the same logic: double down on the tech (chips, AI, robots) before 2030. And the AI Wars podcast series tells that race as an industrial thriller.

In five years — what's already begun

The start of the change is right here. Not in the future. In last quarter's numbers.

Early this year, U.S. tech companies laid off nearly 80,000 people in three months (consolidated figures on Layoffs.fyi, the reference tracker). Nearly half of those layoffs are officially attributed to AI by the companies themselves. Not "we're restructuring." No. Explicitly: "we have tools that do the work, we don't need as many people anymore."

Block, one of the big U.S. payment companies (Jack Dorsey's former Square), went from 10,000 employees to fewer than 6,000 in a few months. Their official reason: their AI agents handle most customer requests with no human involvement.

But the number that struck me most is somewhere else. In the U.S., hiring of young developers dropped 20% in two years. Entry-level tech job postings have been cut to a third. Unemployment among young computer science graduates is at its highest level in twenty years.

Put another way: AIs are already good enough to absorb the tasks juniors used to do. Seniors see their productivity multiplied. Bosses no longer hire juniors because they're expensive to train for a payoff that has vanished.

The problem is that in ten years, we'll be missing a generation of seniors. Because the seniors of 2035 were the juniors of 2025. And those juniors, we didn't train them.

We're sawing off the branch the whole industry is sitting on. Silently.

What's moving in real life

In the streets of major U.S. cities, driverless cars have become ordinary. Waymo, the Alphabet (Google) subsidiary, runs 500,000 trips a week. The service operates in San Francisco, Phoenix, Los Angeles, Austin, and several other cities. London launches this year, the first city outside the U.S.

Meta's smart glasses — the ones that film, answer your questions, recognize what you're looking at — sold 7 million units last year. Apple releases its model at the end of this year.

And humanoid robots, those human-shaped robots that can pick up a cup of coffee or fold laundry, have gone from a sci-fi dream to a product you can buy. The simplest Chinese model, the Unitree G1, costs $5,000. Available right now. Tesla wants to produce tens of thousands this year with its Optimus robot.

I'm not saying these machines will fill every living room by 2031. But the curve is under way. In ten years, a humanoid robot at home will be about as ordinary as a smartphone is today.

In ten years — the great fork in the road

At ten years out, the scenarios split apart. And here I have to be honest: no one really knows.

The optimistic scenario

On one side, the heads of the big U.S. AI companies — Sam Altman of OpenAI, Dario Amodei of Anthropic, Elon Musk of xAI — all say the same thing: within the next two or three years, AI will reach the level of an average human on just about every front. Then it'll surpass them. After that, everything accelerates. Scientific discoveries that take decades today will happen in years. Diseases will retreat. The global economy will double. We'll live longer, in better health.

Amodei goes further in his public appearances: he openly talks about a "compression of a career", an entire professional life that could condense into a few years once AI is powerful enough to do a PhD's worth of work in a few hours. He wrote a piece he calls Machines of Loving Grace. In it he describes what's possible if everything goes well. It's an optimism that makes your head spin.

The opposite scenario

On the other side, researchers who are just as serious argue: we're on the wrong track. Yann LeCun, one of the three pioneers of modern AI (2018 Turing Award), left his role as chief AI scientist at Meta in 2025. His message: current techniques lead nowhere. They're brilliant at imitating, mediocre at understanding. To go further, we'll have to reinvent the research from scratch. That could take ten years, twenty years, maybe more.

Gary Marcus, professor emeritus at NYU and author of Rebooting AI, thinks the financial bubble around AI will burst within the next two years. Hundreds of billions have been invested betting on miracles. If the miracles don't come, the correction will be brutal.

What happens either way

What strikes me is that in both scenarios, three things are more or less certain:

The first is a concentration of power like we've never seen. Five U.S. companies and one Chinese company already share most of the technology. A handful of billionaires — Musk, Zuckerberg, Bezos, Altman — carry more weight than most nation-states. The rules that govern these companies are still being written, and most governments are largely outpaced.

The second is the strain on energy. The data centers that run the AIs consume enormous amounts of electricity. Their consumption is set to double within five years. Microsoft, Google and Amazon are already ordering small nuclear reactors to power their servers. Nuclear fusion — the technology we've been promised for fifty years — is finally reaching the industrial stage. The first deliveries are expected for 2028.

The third is the growing difficulty of knowing what's true. Videos, voices, images faked by AI are becoming impossible to tell apart from the real thing. The 2028 elections, in the U.S. and elsewhere, will play out in an information landscape where everyone will be suspicious of everything.

In twenty years — my son's world

Twenty years is both far off and close. It's the moment my son will finish his studies. It's also roughly the moment I'll turn sixty.

I'm not going to hand you a list of possible scenarios. At twenty years out, no one knows. Instead I'll tell you what I really think, while accepting that I could be wrong.

He'll never drive — 70% odds

My son will never drive. I'm not saying "he won't get his license" — I'm saying he'll never drive. I put 70% odds that he'll never take the driving test, and 80% that he'll never own a personal car. He'll turn 18 in 2040. By then, self-driving cars will be the norm in cities — surely not everywhere, but enough that driving yourself becomes a strange act. The way today, no one learns to light a fire without a lighter anymore. Technically possible, practically useless.

Why not 100%? Because France isn't San Francisco. Rural areas, small roads, mountain trips will probably stay hand-driven for a long time. My son might live in Paris, Lyon or Bordeaux — he won't need a license. If he lives in a village in the Cantal, maybe he will.

He won't go to a human doctor — 60 to 80% odds

When he has a stomachache, or a health question, I don't think he'll book an appointment with a human doctor. Not because human doctors will be banned. Because he'll prefer the AI. I put 60 to 80% odds that within 15 years, more than 50% of primary medical consultations (stomachache, fever, health advice) will happen with an AI — and no longer with a human.

Picture a doctor who has read the entire body of global scientific literature, who's up to date on the latest studies published this morning, who has access to your complete medical record since birth, who could cross-reference your symptoms with those of ten million other patients to tell you precisely what's going on, and who could answer you at three in the morning without getting impatient. That intelligence will exist. And it'll be accessible from your phone.

Who's going to want, instead, to wait three weeks for an appointment with someone who's seen forty patients that day, who can make mistakes because they're tired, and who'll see you for fifteen minutes before moving on to the next one? I think people will prefer the machine. First by choice. Then as the norm. Human doctors will become companions for complex cases, or emotional guides. Not the first point of contact.

Why we'll prefer the machine — the real contrarian thesis

This is the hardest angle to hear, and the most important in this article. I'm not saying AI will replace us by force. I'm saying that we humans are going to actively choose it over other humans. And it's already showing in 2026, in three very concrete areas.

First example: Doctolib + Gemini Health AI. Doctolib has already rolled out an AI assistant that pre-qualifies your consultation before you even see the doctor. You describe your symptom, the AI asks the right questions, points you to the right specialty, offers a slot. Many users already say they prefer this step to the standard exchange with the medical front desk. Less judgment, no feeling of being a bother, sharper questions.

Second example: Waze vs. a human guide. Nobody asks a pedestrian for directions anymore. We prefer Waze. Not because Waze is mandatory — because Waze is better. More up to date, more precise, faster. The shift happened in less than ten years, with no law, no obligation. By preference.

Third example: Khanmigo in pilot schools. Since 2024, the Khan Academy has been testing an AI tutor that works with each student individually. First feedback: many students prefer to ask their "dumb" questions to the AI rather than to the teacher in front of the class. No shame, no judgment, infinite patience.

The psychological mechanism (why we prefer it)

It's not just laziness. It's several forces stacking up:

• Available 24/7 — the AI answers at 3 a.m., on Sundays, when you really need it.
• Zero judgment — you can ask the stupid question you don't dare ask a human (financial, medical, sexual, legal). The AI doesn't raise an eyebrow.
• A complete database — it has read more than any human could ever read in a whole lifetime.
• No day off — no fatigue, no bad day, no bias from being tired at the end of a consultation.
• Total personalization — it knows your record, your tastes, your history. A doctor who sees you for the 3rd time in 10 years can't compete.

The 20% where humans stay irreplaceable

I don't think AI replaces everything. There'll be a 15 to 25% zone where the human stays central, even indispensable:

• Embodied empathy — delivering a serious diagnosis, sitting with someone in grief, holding a trembling hand. The physical presence of another human has a value the machine won't reproduce.
• Ethical judgment in gray areas — end-of-life decisions, complex medical trade-offs, choices that have no right answer.
• Political decision-making — democratic legitimacy, public accountability, the ability to be removed from office. An algorithm can't be elected.
• Live art and sports — a concert with a musician, a game with athletes, a stage play. The human presence is the product.

But for the remaining 75-80% (primary consultation, legal advice, tutoring, guidance, basic financial advice, language lessons), I honestly think we'll choose the machine. Not out of obligation. By preference.

His teachers won't all be human — 80% likely

At school, I think my son will have human teachers for the classes where the human relationship truly matters — sports, the arts, philosophy, group projects. For everything else, he'll have an AI tutor. A tutor that adapts to his pace, that has infinite patience, that knows when he doesn't understand, that finds ten different ways to explain the same thing until it clicks.

We've known since the 1980s that children with a personal tutor learn twice as well as those in a class of thirty. But we could never afford it at scale — the cost was prohibitive. AI changes that equation. One AI tutor for every child becomes possible, at a cost trending toward zero.

I put 80% odds that by 2035, kids in private schools will have a personal AI tutor for most academic subjects, and 50% for public schools (where institutional resistance will be stronger). Parents who have the choice will prefer this option. Again: by preference, not by obligation.

Work won't be what we know

And then there's the question closest to my heart. When my son is 24 and starting his working life, what will the world of work look like?

I don't think it'll look anything like what we know.

I'm not saying there'll be no more work. I'm saying that the nature of work is going to change fundamentally. A large share of the economic value will be produced by machines — AI agents, robots, automated systems. Humans will contribute through what machines can't do: giving meaning, creating genuine novelty, weaving relationships, deciding when the rules aren't enough.

Many countries will set up some form of basic income. Not out of ideology, out of necessity. Because if the majority of wealth is produced without humans, that wealth has to be redistributed somehow — without it, the whole structure collapses.

And the word "work" itself will change meaning. For my son, working will maybe look more like what we call today a project, a commitment, a passion. A choice you make because you want to contribute to something. Not an obligation to survive.

I'm not saying it'll be easy. A society without work in the classic sense is also a society that has to reinvent the meaning it gives to life. That's a huge challenge. But it isn't necessarily a nightmare.

— My bet, with numbers

My 5 bets for 2030-2050

To really own this article as an opinion, I'd rather put numbers down than make vague claims. Here are my 5 concrete bets, with a probability I commit to revising publicly every year. If I'm wrong, I'll own it. If I was right, I'll say that too.

The 5 bets at a glance, before the detail below.

Bet	Horizon	Probability	If true
AI as front-line doctor	2030	60%	> 50% of primary consultations go through an AI before the human doctor.
Self-driving cars mainstream	2032	70%	Driverless service used daily by > 30% of city dwellers in France's 4 biggest metro areas.
1:1 school with an AI tutor	2035	50%	50% of upper-middle-class families pay for an AI tutor on top of school. Widening inequality.
Driver's license optional	2040	80%	The license is no longer a mandatory social milestone in metro areas. My son may never drive.
AI makes > 30% of strategic decisions	2050	65%	The human CEO becomes a signatory / ethical guarantor, no longer an operational decision-maker.

1. 2030 — AI doctor (60% odds) · More than 50% of my primary consultations (sore throat, fever, health advice) will go through an AI before any human contact. The human doctor will be consulted as a second line, on complex cases or for procedures (vaccine, physical exam). Reasoning: Doctolib + Gemini Health are moving fast, the shortage of GPs in France creates the demand, and user convenience will do the rest.
2. 2032 — Self-driving cars mainstream (70%) · In every major French metro area (Paris, Lyon, Marseille, Toulouse), a driverless car service will be operational and used daily by more than 30% of city dwellers. Waymo is 5 years ahead in the U.S., Europe is catching up with a 4-6 year lag, French regulation drags but always caves to convenience.
3. 2035 — 1:1 school with an AI tutor (50%) · Half of upper-middle-class families will pay for a personalized AI tutor for their kids on top of school. It won't be a public revolution — it'll be a silent deepening of educational inequality between equipped and unequipped families. Medium probability because resistance from teachers' unions and the up-front cost of good AI tutors will slow adoption.
4. 2040 — Driver's license optional (80%) · By that date, my son will be 18. I put 80% odds that he lives in a metro area where the driver's license is no longer a socially mandatory milestone. Just as today no one is surprised that a 30-year-old Parisian has never driven, that'll be the norm in every big city.
5. 2050 — AI makes > 30% of strategic corporate decisions (65%) · More than a third of leadership decisions (hiring, budget allocation, investment choices, dynamic pricing, supply chain optimization) will be made by AI systems with pro-forma human validation — or with no validation at all in the most advanced companies. The human CEO will remain, but as a signatory and an ethical guarantor, not as an operational decision-maker.

What I want to teach him

Everything I've just written is my opinion. It's worth what it's worth. What really matters is what I do now with these four years I have with my son before he starts primary school.

What do I teach him?

Not to code. AIs code better than me. In fifteen years, they'll code better than anyone. It's no longer a good bet for a child.

Not to memorize dates or capitals. He'll always have, in his pocket, a machine that knows everything and answers in a second.

Not to do calculations by hand. Same point.

So what?

I want to teach him to judge. To sense when something rings false. Not to believe everything he sees. To ask the right questions. Because in his world, information will be everywhere, but discernment will be rare.

I want to teach him to choose. Not to just take what he's offered. To understand why he prefers this over that. To own his tastes, his values, his disagreements.

I want to teach him to love. To be in a real relationship with other humans. Because the human relationship, authentic, with its silences and its awkwardness, is maybe the one thing machines will never truly do.

I want to teach him to create. Not to copy, not to follow. To invent something that didn't exist before. Even small, even clumsy. The joy of bringing something new into being.

And then, paradoxically, I want to teach him not to be afraid of the machine. To understand how it works. To steer it rather than be steered by it. To stay in a position of choice, at the moment when everyone around him will prefer the choices already made for them.

Going further

If the subject interests you, three reads that fed my thinking:

1. Machines of Loving Grace — Dario Amodei — the most important essay on the optimistic scenario, written by the head of Anthropic.
2. The Precipice — Toby Ord — an Oxford philosopher who thinks AI is the greatest risk humanity has ever faced.
3. Situational Awareness — Leopold Aschenbrenner — a former researcher at OpenAI who describes, unfiltered, what the labs see coming.

And three direct sources to track the bets in real time:

4. Khan Labs · Khanmigo — the AI tutor in pilot schools, early public feedback on 1:1 learning with AI.
5. Doctolib blog — to follow the concrete progress of AI assistants in the French medical journey.
6. Waymo blog — quarterly figures on the self-driving car rollout (San Francisco, Phoenix, Los Angeles, Austin, London from 2026).

To dig into the complementary angles on this site, also read my breakdown of China's 2026-2030 plan (the geopolitical dimension of the same AI race), the AI Wars podcast (the OpenAI vs. Anthropic vs. Chinese labs rivalry as a narrative series), and Karpathy's work (to understand, technically, where the models stand).

This article is an opinion. It'll evolve with the facts. If you want to follow how it evolves, sign up for my newsletter. And if you don't agree, if you think I'm too pessimistic or, on the contrary, that I'm sugarcoating it, reply to the email, I read everything. It's too important a subject for me to talk about it alone.

— FAQ

The world and AI, FAQ.

What will the world look like in 5 years with AI?

My bet: the tech job market will already be broken. In early 2026, U.S. tech companies laid off nearly 80,000 people in a single quarter, half of them explicitly because of AI. Junior developer hiring dropped 20% in two years. Driverless cars already run 500,000 trips a week in the U.S. Within 5 years, what's marginal today will be the norm.

Will my 4-year-old son ever need to learn to drive?

My personal bet: no, he'll never drive. He'll turn 18 in 2040, and by then self-driving cars will be the norm in cities. Driving yourself will become a strange act, like lighting a fire without a lighter. Technically possible, practically useless.

Will AI replace my job within 10 years?

I think we're looking in the wrong direction when we talk about replacement. The real shift is that we ourselves will choose the machine over a human, because it'll be more available, more up to date, more patient. Researchers measure that an AI's autonomy doubles every 4 months. Projected over 10 years, that changes the very nature of work, not just jobs.

Which jobs are most exposed to AI today?

Career entry points, especially in tech. Job postings for junior developers in the U.S. have dropped by two-thirds and unemployment among computer science graduates is at a 20-year high. Block went from 10,000 to under 6,000 employees thanks to its AI agents. The real risk isn't senior staff disappearing, it's the absence of juniors we no longer train, and therefore the disappearance of tomorrow's seniors.

Should we fear AI taking over?

My take: the real threat isn't Skynet, it's the concentration of power. Five U.S. companies and one Chinese company share most of the AI tech. A handful of billionaires (Musk, Zuckerberg, Bezos, Altman) carry more weight than most nation-states. Governments are largely outpaced. The risk is that imbalance, not a sci-fi movie.

How do you prepare your kids for a world dominated by AI?

My choice for my son: don't code, don't memorize, don't calculate by hand. Teach him to judge (to sense when something rings false), to choose (to own his tastes), to love (real relationships) and to create (to invent what didn't exist before). And teach him to steer the machine rather than be steered by it. Deep roots so he can truly inhabit his world.

Why do you think people will prefer an AI doctor to a human one?

Picture a doctor who has read the entire body of global scientific literature, who's up to date on studies published this morning, who has your complete medical record since birth and who'd answer at 3 a.m. without getting impatient. Against that, waiting 3 weeks for a 15-minute appointment with someone tired who has seen 40 patients that day? My bet: people will prefer the machine. First by choice, then as the norm.

Is the scenario of an AI bubble bursting credible?

Credible, yes. Gary Marcus, a critical researcher, believes the financial bubble around AI will burst within 2 years. Hundreds of billions have been invested betting on miracles; if the miracles don't come, the correction will be brutal. Yann LeCun, an AI pioneer, left Meta in 2025 saying current techniques lead nowhere. Even in that scenario, the concentration of power and the energy strain are here to stay.

Will my kids' school change with AI?

My bet: yes, deeply. We've known since the 1980s that a child with a personal tutor learns twice as well as in a class of 30. Cost made that impossible at scale. AI changes the equation: one AI tutor per child at a cost trending toward zero. Human teachers will stay for sports, the arts, philosophy, group work. The rest will shift, driven by parents' preference, not by force.

How do you live calmly with these AI scenarios?

Three reads that helped me find my footing: Machines of Loving Grace by Dario Amodei (Anthropic) for the optimistic scenario, The Precipice by Toby Ord (Oxford) for the existential-risk bet, and Situational Awareness by Leopold Aschenbrenner (ex-OpenAI) for what the labs are seeing. I can be wrong on many points, but I think we're underestimating what's coming. Read, doubt, choose.