Where Do People Go from Here?

Let us stop arguing for a moment about whether today’s artificial intelligence is good enough.

Suppose that five years from now it can reliably perform a significant share of intellectual tasks faster, more cheaply and, on average, better than a person. Not every task. Not work as a whole. But enough of it for this to become ordinary economics rather than a capabilities demonstration.

Fine.

Where do people go from here?

Five years is a scenario, not a date in the diary

I do not know whether such a transition will happen by 2031. But it is no longer honest to put the entire horizon in the science-fiction category. METR’s research has shown rapid growth in the duration of tasks advanced agents can complete autonomously, while the OECD is already considering markedly different trajectories for AI capabilities through 2030.

Labour-market evidence today is calmer than the headlines. The ILO expects primarily a transformation in the content of work rather than the sudden disappearance of whole occupations. Across the macro trends it models, the World Economic Forum sees tens of millions of roles displaced — and still more created.

That describes the transition. I am interested in the next frame: what if producing the same amount really does require substantially fewer people?

“New jobs will appear” does not solve the arithmetic

New occupations did emerge after previous technological shifts. They probably will this time as well.

But there is no law of nature requiring nine vanished jobs to be replaced by nine new ones — with the right skills, comparable pay and, preferably, somewhere near home.

If one professional using AI can produce what previously required a team, some people will move into more interesting work. Some will switch sectors. And some will simply discover that the market no longer needs the previous volume of human labour.

That is not necessarily a catastrophe. But it is very different from the comforting story in which everyone freed from routine work immediately becomes a strategist and spends nine to five thinking creatively.

Nature is under no obligation to create a new occupation with the same number of vacancies for every one that disappears.

The physical world stages a comeback

While intelligence becomes cheaper inside computers, material reality remains stubborn. A pipe still has to be replaced. A patient has to be physically moved. A house has to be built. A child has to be collected from nursery. Robotics will advance here too, but hardware, energy, safety and maintenance move more slowly than pure software intelligence.

We may enter a rather amusing period in which a good analyst finds it harder to defend their profession from automation than an electrician does.

For decades we have told children: “Learn to work with your head.” Parents may eventually start grumbling: “Why do you need all that data analytics? Learn a proper trade. At least know how to wire a house.”

History enjoys reversals like that — particularly once the previous generation’s certainty has already made it into the curriculum.

The human being as part of the service

Some work may survive even after machines become better at the formal function.

Not because AI remains technically behind. Some products simply include a human being as part of what people value.

People want to see a doctor, speak to a teacher, trust a carer, argue with a live negotiator, listen to a performer who has actually lived through what they are singing about. Not always — but often enough for human presence itself to become part of the product.

One day, “served by a real person” may become the equivalent of “handmade”. The mass service is automated; the live professional is the premium option.

After decades of fighting contact centres, humanity may finally achieve the right to speak to an operator. Possibly at a higher tariff.

The hardest question is not what people will do, but how the result is distributed

Modern economies link production and distribution through work. Most people sell time and expertise, receive income, and use that income to buy a share of what society produces.

If the same level of output requires less labour, nothing automatically breaks. But the familiar arrangement begins to creak.

We can shorten the working week. We can redistribute through taxation. We can broaden capital ownership. We can create new services and new human wants. We can do all of these at once in different proportions and produce several very different societies.

At this point, the conversation about AI stops being a conversation about technology. A machine may increase output. It does not determine who should own the gains from higher productivity.

That is why “where do people go?” eventually becomes something other than a question about the geography of new professions. It becomes a question about the role labour plays in distributing money, time and status.

If less human labour is required, this is not only an employment problem. It is a question of how the gains are distributed.

Perhaps part of the answer is simply to work less

There is one option that is oddly often described as defeat: if production requires fewer human hours, we could in fact spend fewer human hours producing things.

For centuries we have created tools to free ourselves from heavy and repetitive work. Then, just as they succeed, we become anxious about how to fill the eight-hour day they have released.

People will still start companies, compete, explore, argue, build, learn and do things whose necessity cannot be proved in a financial model. We have plenty of experience with that.

But perhaps the equation “a decent life = compulsory full-time employment” will one day stop feeling quite so natural.

If that happens, the argument over whether AI replaced a person in one particular role will seem rather small. The question will no longer be where people are supposed to work, but how much work a person actually needs for a decent life.

Sources and reference points

The factual reference points below are current as of September 2026. The future scenarios in the article are the author’s assumptions, not forecasts made by these organisations.

1. ILO / NASK, Generative AI and Jobs: A 2025 Update. Source (https://www.ilo.org/publications/generative-ai-and-jobs-2025-update)
2. World Economic Forum, Future of Jobs Report 2025. Source (https://www.weforum.org/publications/the-future-of-jobs-report-2025/in-full/4-workforce-strategies/)
3. METR, Measuring AI Ability to Complete Long Tasks. Source (https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/)
4. OECD, Exploring possible AI trajectories through 2030. Source (https://www.oecd.org/en/publications/exploring-possible-ai-trajectories-through-2030_cb41117a-en.html)