Almost nobody works voluntarily. That is neither an accusation nor an outburst, just what you see when you ask: people go to work because the alternative is not getting by. Economic compulsion, in other words. We usually call it something else.

It matters, since being needed is normally counted among the things we stand to lose when human labour stops being necessary. But most of what we call being needed at work is not being needed. If you fail to turn up for a shift, someone can cover — a stand-in, a reshuffle, at worst a delay. What genuinely cannot be replaced is something else: a child, a friend who calls at two in the morning, a team that is one man short, a neighbour who cannot manage if nobody looks in. There is no hire to put in that gap.

The objection sharpens the argument rather than weakening it. If what is lifted away when economic compulsion ends is mostly compulsion, then it is not meaning that disappears. What disappears is an imitation — a form of being needed that resembled the real thing closely enough that we never noticed we lacked it. The danger is therefore not that we lose something we had, but that we discover we never built it.

One qualification straight away, or it will chafe throughout. Being needed is one source of meaning among several. The joy of making things, curiosity, belonging, beauty, love, mastery — none of it disappears when work does. What makes being needed worth an essay is that it is the only one of them that a society without economic compulsion systematically withdraws. That asymmetry is the point; the rest of the list I mention and move past.

What grew when Swedes got time

Between 1992 and 2019 the share of adult Swedes who had done voluntary work in the past year was essentially unchanged: between 48 and 53 per cent, survey after survey.1 Twenty-seven years of rising real incomes and better health — and the curve barely moves.

But something did move in the same data. Informal help — regular support for someone outside your own household, unpaid or for token compensation — rose from 28 to 46 per cent over the same period.1

One curve measures organised activity. The other measures someone being personally dependent on you. When Swedes got more time, more money and better health, it was not the activity that grew. It was the dependence.

The increase should be read with reservations, and there are several. The series is not a smooth rise: it is flat through the 1990s, jumps to 52 per cent in 2005, falls back and rises again towards the end. The report itself notes that part of the jump follows from the question being widened in 2005 to capture companionship, looking in and minding. The population also aged over the period, and the report's own age breakdown shows the sharpest increase in the 75–84 group. How much of the rise is demography, how much is the change in wording and how much is real change cannot be settled from the material. The direction holds nonetheless.

Two kinds of occupation

Meaningful occupations fall into two kinds with entirely different mechanics.

The first lies in the activity itself. You play in order to play, read in order to read, climb the mountain because the mountain is there. The meaning is built into the act and requires no recipient.

The second does not lie in the activity at all, but in the fact that someone else is worse off if you do not do it. The parent. The person sitting with the dying. The fourth crew member of a retained fire unit called out at three in the morning. What the act consists of matters less; what carries it is the dependence.

The difference shows when you remove the recipient. In the first kind: nothing happens. In the second: the activity collapses immediately. Nobody sits in a waiting room for the waiting room's sake.

The distinction can be tested, because we already have a large-scale case in which economic compulsion ends: retirement. If what mattered lay in the activity, the outcome ought to be uniform — everyone gets the same thing, namely time. That is not what the research shows. A meta-analysis of more than half a million individuals finds protection rather than harm, just under 20 per cent lower risk of depression, but with high heterogeneity.2 An older American panel analysis points the other way: a 6–9 per cent deterioration in mental health on full retirement.3 European longitudinal data show a short-term improvement that does not merely fade but reverses: the risk of depression rises again from the tenth year in some groups, and the risk of suicidal ideation is 30 per cent higher after five to nine years and 47 per cent higher after ten years or more.4 Japanese data on 62,438 people show increased depressive symptoms, most clearly among men from lower occupational classes.5

The figures do not point the same way, and that is itself the result: the dispersion is larger than the mean. What explains it is harder to establish than one would wish. Social participation — roles, assignments, settings where someone counts on you — is the factor singled out, but the support is weaker than the argument deserves: in the study that tested the question most directly, the mitigating effect was a non-significant trend, and it appeared only for recreational participation, not for volunteering.5 And the most common role is the most prosaic one: between 24 and 60 per cent of those who are grandparents look after grandchildren.6

The analogy is incomplete, and that should be said plainly. Someone retiring has a working life behind them and a transition that is culturally legitimate in a way that an imposed one never is. The analogy therefore probably overstates how well things go. That makes the conclusion more cautious, not weaker: even under the most favourable conditions we know of, the outcome of freed-up time is a distribution, not a state, and what decides where you land is whether anyone needs you.

The scarcity we choose

If dependence is what carries, people ought to manufacture it where it is absent. They do.

For the Vasaloppet ski race on 1 March 2026, 14,300 people had entered to ski 90 kilometres between Sälen and Mora. Counting the whole of the race week, the figure was more than 61,600. Almost half of those entered for the main race had never skied it before.7 Nobody depends on their doing so. Nothing is produced, nobody is supported, and transport between the two towns was long ago solved by car in under two hours.

What is going on is that the distance is being put back deliberately. You remove the engine, reinstate the stretch and make it hard again. The same mechanism appears in crafts performed more slowly than the machine could manage, in gardens that cost more than the vegetables, in choirs rehearsing for six months for a concert nobody asked for.

I have used the term artificial scarcity before, and always negatively: platforms throttling access to something that is free to copy, subscriptions turning ownership into rental. That is scarcity someone else imposes on you in order to charge for it. Here the same mechanism is turned the right way round.

What separates it from mere enjoyment is that chosen scarcity creates an obligation. A start time to be kept. A handover where someone stands waiting. A team that is one man short. The construction is voluntary; the obligation inside it is not, and it is the obligation that does the work. Chosen scarcity is not play. It is manufactured dependence.

The mechanism is fussy. It works only when the scarcity is real once chosen. The track is real, the cold is real, and whoever has not trained will be left standing at Evertsberg. A constructed necessity whose resistance can be negotiated away on the spot is no necessity. It is a competitive discipline in self-deception.

Being needed cannot be printed

If the argument holds, an uncomfortable problem follows, and it is my own.

Calories can be produced in greater quantity. Dependence cannot. Being needed presupposes someone who lacks what you have. A mentor requires someone who cannot yet do it. A caring assignment requires a care need. An emergency call requires an accident. The better a society becomes at abolishing dependencies — and abolishing dependencies is more or less the definition of welfare and technology — the fewer remain to distribute.

This is a structural constraint, not a question of status. Some roles cannot be given to everyone at once, because they presuppose that someone stands at the opposite end. The patron needs someone to fund, the mentor someone to train. If everyone is a patron, nobody is left to give to. Access to education, healthcare and calories can be expanded; access to someone needing you specifically cannot.

Here lies the labour market's most underrated function. Wage labour was a poor distribution mechanism for money and a strangely effective distribution mechanism for dependence. It placed millions of people daily in situations where someone else depended on their turning up. That most of this was compulsion rather than genuine need does not make it insignificant — the compulsion produced real being-needed as a by-product, on a scale nothing else we have built comes close to.

That is the function that disappears first, long before the wage does.

Those who never get to start

I have written before about what happens when the development ladder is pulled up for entire countries: the industrial route and the services route can vanish simultaneously for economies that still need them.8 I will not repeat that analysis. But the movement has an individual counterpart that I left untreated.

Anyone who never enters working life never gets even the counterfeit version of being needed. That sounds like a small loss, given that the version is counterfeit. It is not. Counterfeit being-needed comes with a standing — a shift, a colleague who counts on you, a place where your absence is noticed — and it is out of that standing that genuine dependencies tend to grow. You become the one people ask for help, you end up on the association's board, you acquire a reputation for being reliable. In practice, the route to the genuine has run through the counterfeit.

The research shows narrower things than it is usually used for. That early exclusion is costly is well established: Swedish vocational upper-secondary graduates who were unemployed straight after finishing had roughly 17 per cent lower annual income five years later than their siblings.9 Belgian school-leavers still unemployed after nine months had dramatically worse odds of getting in at all — if entry was delayed by a further year, the probability of finding work within two years fell from 60 to 16 per cent for men.10 But the literature is not unambiguous in the direction it is usually cited for. The most comprehensive review finds that the income effects of entering during a recession fade after ten to fifteen years — while at the same time reporting lasting effects on family formation, health and mid-life mortality, and an earnings gap that reopens at precisely that age.11 And the study most often invoked as proof of permanent scarring in fact draws the opposite conclusion about employment.12 The picture, then, is not that the scars are smaller than assumed. It is that they sit somewhere other than where the literature tends to look.

Closest to my argument lies not the income research but the health research. Everyone who finished the final year of compulsory school in a medium-sized industrial town in northern Sweden in 1981 — Luleå — was followed for twenty-seven years, with mental health measured already at sixteen, that is, before exposure. Open youth unemployment was associated with an elevated risk of mental health problems at twenty-one and still at forty-three: odds ratios of 2.48 and 1.71 respectively. Labour market programmes were not — 0.95 at twenty-one and 1.23 at forty-three, the latter not statistically significant.13

The difference between the two is not the income. It is that someone expected you to turn up.

I do not conclude that it was being needed that did the work. The programmes also provided structure, activity and belonging, and the study cannot separate the components. The authors themselves write that the programmes leave "no or minimal scars" — not none at all; in the unadjusted model, programme participation was in fact linked to worse health at forty-three. The cohort is moreover a single town, at a time when adult unemployment stood at around three per cent. But it is the most telling contrast I have found, and it points the same way as everything else in this essay.

A separate analysis of the same cohort shows the asymmetry across the life course: youth unemployment is associated with worse mental health at 21, 30 and 42, while unemployment later in life produces problems at the time but not lasting ones.14 That is exactly what one would expect if what is damaged is entry into a role rather than an income. But the studies are conducted in a single industrial town and cannot separate cause from selection.

The present situation calls for caution. Youth unemployment in Sweden averaged 24.3 per cent over 2025 — a figure that looks worse than it is. Unemployment is measured as a share of the labour force, and most people in that age group study full time without seeking work and are therefore not counted in the labour force at all. Full-time students who do seek work, on the other hand, are counted as unemployed: 102,000 of the 170,000 unemployed young people were full-time students.15 In American payroll data, employment among 22–25-year-olds in the most AI-exposed occupations is 19 per cent below the level it would have reached had it tracked peers in less exposed occupations, and the mechanism is absent hiring rather than dismissal. The authors state explicitly that they do not show AI caused it.16

Established, then: early exclusion costs in income, and it costs in mental health in a way that later unemployment does not. Unestablished — my own inference — is that the latter cost arises because what is missing is not a wage but an entry.

Cultural asymmetry

If what has to be replaced is a distribution mechanism for dependence, the shift lands with different force depending on where in a society the chains of dependence sit.

Societies that never located human worth in productivity have less to lose. Where standing derives from lineage, age, religious position or family, the chains of dependence already sit outside the labour market and keep working as it thins out. That is not an argument that such arrangements are better, but a statement about what they are exposed to.

The secular, individualistic achievement cultures are the most exposed, and Sweden is among them. The Swedish work-first doctrine is not an administrative arrangement but a moral idea with deep roots in Lutheran northern Europe: whoever works contributes, whoever draws a wage has earned it, whoever stands outside becomes not only economically but symbolically weaker.17 Such an idea does not disappear when work does. It becomes homeless, looks for something to attach to, and readily attaches to the wrong things.

It is the same kind of inversion that applies demographically: where the workforce is shrinking, technology solves a shortage; where it is growing, technology removes the entry-level jobs. The risk is therefore greatest where the need for labour is greatest.8 The more tightly a society has tied human worth to productivity, the greater the loss when productivity becomes free.

The closest real cases are the rentier economies of the Gulf, where material scarcity is largely alleviated by resource income. Around three-quarters of employed Kuwaitis were working in the government sector at the end of 2025.18 What that has done to the experience of meaning I do not know, and will not pretend otherwise. What I can say is weaker: where the work requirement was lifted, no new way of being needed was built, but public employment that the IMF describes as oversized, costly and a drag on productivity growth in the economy as a whole. Jobs were manufactured instead of dependencies. A warning sign, not a proof.

Where it hurts

Two points about the sequence in time, both offered as premises rather than investigations.

The first: the distribution of material abundance is a political question, not a technical one. The world's supply of dietary energy passed 3,000 kilocalories per person per day in 2023 — 3,016 kilocalories, far above average global energy requirements, albeit measured as availability at the retail level rather than actual intake19 — and yet between 638 and 720 million people were undernourished in 2024.20 Abundance did not abolish scarcity, because hunger was never a production problem.21

The second: the crisis comes in the transition, not at the destination. The cognition curve and the atom curve are diverging — the cognition-heavy accelerates while robotics, power grids and construction lag.22 Anthropic's August risk report supports the picture, with the limits of its method openly stated. Section 3.6 rests on semi-structured interviews with 31 experts, described by Anthropic themselves as "a rough litmus test, rather than a rigorous or comprehensive assessment". One of the five energy experts interviewed emphasised that "physical deployment (e.g. building transmission infrastructure), not R&D, is a significant and not-yet automated constraint". At the same time, Anthropic writes in its own name that it is plausible that automated research and development will become "a major concern in the next 6–12 months".23

Individual judgements, not measurements. But read together they give the transition a shape: cognition may run ahead while material abundance queues for transformers and permits. An occupation can lose its value long before prosperity reaches the person who practised it. It is in that gap that the question bites: what am I when what I could do is no longer in demand, and the money has not yet arrived?

What is weak in this

Four objections to my own argument, in descending order of how much they trouble me.

The mixing problem. Compulsion and genuine being-needed cannot be cleanly separated, and the entire essay presupposes that they can. The nurse works for the wage and is genuinely needed, simultaneously, in the same act. The shift is economically imposed and the patient genuinely dependent. That I can point to pure cases at either end does not mean the middle is empty; the middle is probably where most people are. Worse: compulsion has produced a great deal of real being-needed as a by-product. It forced people into rooms where others actually needed them. That this being-needed will arise by itself once the compulsion is removed is an assumption, and I have no support for it.

The transparency problem. Manufactured necessity works only if you can forget it was manufactured. The ski track carries because the cold and the body are real. A civic assignment that everyone knows is arranged for the participant's sake risks feeling like therapy. I see no obvious route from "we ought to construct necessity" to "this feels necessary", and that is precisely the route that would be needed.

The availability problem. Voluntary scarcity is already available to everyone — associations, choral singing, crafts, retained fire services, endurance sport — and yet the share who take part in organised form has stood still for twenty-seven years, even as both time and money increased.1 If self-chosen necessity were the solution, we ought to have seen it grow with prosperity. The honest reading is that what grew instead was informal help, that is, dependence itself. But I cannot rule out the simpler explanation: that most people simply do not want to.

Administrability. Distributing goods is routine for a state. How to write the distribution of being needed into a budget bill I do not know. The institutions I have proposed before — learning centres, civic assignments, support for care work24 — distribute opportunities, not dependence. Nobody becomes needed by being assigned a task.

Freeing up is never the point

Let me close with what I know best.

An organisation that frees up capacity without having decided what it is to be directed towards decays. It is the most predictable thing I know of in improvement work. You remove waste, streamline a line, get time back — and if nobody has decided what the time is for, it fills up with something worse than what was removed. Freeing up is never the point. Redirection is.

I see no reason to believe a society behaves better than a workshop. And the redirection required is not primarily about money or time. It is about dependence: building settings where it actually matters whether a particular person turns up or not.

Here the objection I began with lands. If almost nobody works voluntarily, then it is not meaning that is lifted away when economic compulsion is lifted away. It is the imitation — the form of being needed that resembled the real thing closely enough that we never noticed we lacked it. The danger is therefore not that we lose something we had. It is that we discover we never built it.

The question after work is therefore not what we shall do with the freedom. It is whether there is anyone who notices if we do not turn up.

Source notes

Where I go beyond what can be documented, I mark it in the running text.

  1. Johan von Essen and Lars Svedberg (eds.), Medborgerligt engagemang i Sverige 1992–2019, Ersta Sköndal Bräcke University College, working paper series 98, 2020; the chapter on informal contributions is by Magnus Jegermalm. The share of adults (16–84) who had done voluntary work in the past year has stayed between 48 and 53 per cent throughout the series. Regular informal help for someone outside one's own household: 28 per cent in 1992, 30 in 1998, 52 in 2005, 45 in 2009, 41 in 2014, 46 in 2019. The question reads "unpaid, or for token compensation" and is bounded against one's own household, not against organisations. The report itself notes that a widening of the question in 2005 accounts for part of the level shift. Breakdowns by sex, age, donor profile and foreign background are given in the report; the age breakdown shows the sharpest increase in the 75–84 group, from 25 per cent in 2014 to 46 per cent in 2019.
  2. Does retirement trigger depressive symptoms? A systematic review and meta-analysis, Epidemiology and Psychiatric Sciences 30:e77, 2021 (Odone et al., Italian Working Group on Retirement and Health). 557,111 individuals across 60 datasets; pooled effect size 0.83 (95% CI 0.74–0.93), that is, just under 20 per cent lower risk of depression, with pronounced heterogeneity (I² = 93.4 per cent).
  3. Dave, Rashad and Spasojevic, The Effects of Retirement on Physical and Mental Health Outcomes, NBER Working Paper 12123, 2006: a 6–9 per cent decline in mental health on full retirement, and a 5–16 per cent increase in difficulties with mobility and daily activities.
  4. Giansanto Mosconi, Giacomo Pietro Vigezzi, Paola Bertuccio, Andrea Amerio and Anna Odone, Transition to retirement impact on risk of depression and suicidality: results from a longitudinal analysis of the Survey of Health, Ageing and Retirement in Europe (SHARE), Epidemiology and Psychiatric Sciences 32:e29, 2023. 8,998 people across 27 European countries plus Israel, median follow-up nine years. The risk of depression was 11 per cent lower in the first year after retirement (RR 0.89), but rose again from the tenth year among former non-manual workers (RR 1.21) and late retirees (RR 1.37). Suicidal ideation: RR 1.30 after five to nine years and RR 1.47 after ten years or more.
  5. Retirement and mental health: does social participation mitigate the association? A fixed-effects longitudinal analysis, BMC Public Health 17:526, 2017 (Shiba, Kondo, Kondo, Kawachi). Japan Gerontological Evaluation Study, N = 62,438, 2010–2013. Increased depressive symptoms on retirement, most clearly among men from lower occupational classes. The mitigating effect of social participation was a statistically non-significant trend (β = −0.22, 95% CI −0.72–0.29 for men) and applied to recreational participation; for participation with roles, including volunteering, the study found no interaction at all. What the study highlights as the mechanism is a perceived sense of meaning rather than participation as such. Japanese data should not be transferred to Nordic conditions without qualification.
  6. The prevalence of grandparental childcare in Europe: a research update, European Journal of Ageing 20:37, 2023 (Zanasi, Arpino, Bordone, Hank), based on SHARE in 26 countries: the share of grandparents — people aged 50 or over with at least one grandchild — providing any form of childcare ranges from 24 per cent in Latvia to 60 per cent in Belgium and the Netherlands, with an average of 46 per cent. Age-adjusted estimates for 2019–20. The spread is the point, not the level.
  7. Organiser's press material ahead of the 102nd race, 26 February 2026: 14,300 entries for the main race, more than 61,600 entries across the whole of the winter race week, 47 per cent of those entered for the main race had never taken part in the Vasaloppet before. Entry figures, not starters.
  8. My own essay: "When the ladder was pulled up", ai-skiftet.se, 22 June 2026, on development routes being pulled up before more countries have had time to climb, and on the inverted demographic risk.
  9. Oskar Nordström Skans, Scarring effects of the first labour market experience: A sibling based analysis, IFAU Working Paper 2004:14. All Swedish youths graduating from vocational upper-secondary programmes in 1991–94, compared with their siblings: around 17 per cent lower annual income and a 3 percentage point higher risk of unemployment after five years. The figure is a coefficient on log annual earnings without conditioning on employment; it therefore mixes the wage and employment margins and is not a pure wage effect. Two further qualifications: the paper was never published in a peer-reviewed journal, and the cohorts met the labour market in the middle of the Swedish crisis of the 1990s. In a later version (IZA DP 5565, 2011) the author himself describes the effects as persistent but not permanent.
  10. Bart Cockx and Matteo Picchio, Scarring effects of remaining unemployed for long-term unemployed school-leavers, Journal of the Royal Statistical Society Series A, 176:4, 2013, pp. 951–980. The fall from 60 to 16 per cent concerns the probability of finding work within two years, for men whose entry was delayed by a further year. The group is already selected: Belgian school-leavers unemployed for more than nine months, in a system with unlimited benefit duration. The study explicitly finds no direct scar in entry wages or job quality — but an indirect one, via forgone work experience, since each year of experience raises future entry wages by about 2.5 per cent. The lasting scar sits in the probability of getting in at all.
  11. Till von Wachter, The Persistent Effects of Initial Labor Market Conditions for Young Adults and Their Sources, Journal of Economic Perspectives 34:4, 2020. Entering in a typical recession — a rise in unemployment of 3–4 percentage points — produces a 10–15 per cent initial reduction in earnings, an effect that according to the review fades after ten to fifteen years, longer in deeper crises. The review simultaneously reports lasting effects on fertility, family formation, criminal activity, health and mid-life mortality, and notes that the earnings gap reopens in middle age. It concerns economic conditions at entry, not individual unemployment — a different question from note 9, and the two should not be conflated.
  12. David T. Ellwood, Teenage Unemployment: Permanent Scars or Temporary Blemishes?, in Freeman and Wise (eds.), The Youth Labor Market Problem, University of Chicago Press, 1982, chapter 10. Routinely cited as evidence of permanent scarring. Ellwood finds wage effects but states explicitly that his data give no support for time out of work triggering a long-term cycle of recurrent non-employment: "experience dependence yes, but a serious 'permanent scar,' no". The sample is small — around 750 young American men were drawn from the NLS Young Men, but the empirical estimates rest on 271–298 individuals — and Ellwood himself notes that the period had tight labour markets and that his data did not allow him to test how long the wage differential persists.
  13. Mattias Strandh, Karina Nilsson, Madelene Nordlund and Anne Hammarström, Do open youth unemployment and youth programs leave the same mental health scars? — Evidence from a Swedish 27-year cohort study, BMC Public Health 15:1151, 2015. All those who completed, or should have completed, their final year of compulsory school in 1981 at age 16, n = 1,083, 94.3 per cent retained after 27 years counted among those still alive; mental health measured at age 16, that is, before exposure. Open youth unemployment: odds ratio 2.48 (CI 1.57–3.60) at age 21 and 1.71 (CI 1.20–2.43) at 43. Labour market programmes: 0.95 (CI 0.72–1.26) at 21 and 1.23 (CI 0.93–1.63, not statistically significant) at 43. All values refer to the fully adjusted model 4; in the unadjusted model, programme participation was significantly linked to worse mental health at 43 (OR 1.41, CI 1.13–1.65), and the authors summarise the programmes' effect as "no or minimal scars". The article gives the location as "a medium-sized industrial town in northern Sweden"; that it is Luleå follows from the cohort profile (Hammarström and Janlert, Cohort Profile: The Northern Swedish Cohort, International Journal of Epidemiology 41:6, 2012). The study cannot determine which component of the programmes makes the difference; the reading that it is about someone expecting you to turn up is mine.
  14. Mattias Strandh, Anthony Winefield, Karina Nilsson and Anne Hammarström, Unemployment and mental health scarring during the life course, European Journal of Public Health 24:3, 2014, pp. 440–445. The same cohort as note 13, but a different question: unemployment across three life periods rather than unemployment against programmes. Youth unemployment is associated with worse mental health at ages 21, 30 and 42. Unemployment between 21 and 30 was linked to worse mental health at 30 (B = 0.34) but not at 42; unemployment between 30 and 42 was not significantly linked to health at 42. The effects are modest — youth unemployment gives B = 0.28 on an index with a standard deviation of around 1.4–1.8 — while accumulated unemployment across all three periods gives B = 1.28. An observational study without instruments; selection cannot be ruled out as an explanation.
  15. Statistics Sweden, Youth unemployment in Sweden, annual average 2025: 24.3 per cent among 15–24-year-olds (women 23.9, men 24.7). According to the Labour Force Surveys, 170,000 young people aged 15–24 were unemployed in 2025, of whom 102,000 were full-time students. Statistics Sweden notes itself that unemployment is reported as a share of the labour force, and that a large part of the age group studies full time without seeking work and is therefore not counted in the labour force at all. Monthly figures for June are heavily inflated by the end of the school year and should not be used.
  16. Erik Brynjolfsson, Bharat Chandar and Ruyu Chen, Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence, Stanford Digital Economy Lab, working paper, August 2026 version. Employment among 22–25-year-olds in the two most AI-exposed occupational quintiles fell around 11 per cent between November 2022 and June 2026, while it grew around 10 per cent in the three least exposed — a divergence of 19 per cent relative to the latter's trajectory. The authors state explicitly that the work "does not estimate a causal impact of AI" and describe the results as descriptive, note divergent trends predating the generative wave, and note that the pattern attenuates when controlling for education — although they add that the education channel may be the very route through which AI operates. American payroll data from ADP, which also funds the lab and has the right to review the paper before publication; not a representative sample. Earlier versions gave 13 and 16 per cent using a different estimation method — those figures still circulate and are out of date.
  17. My own essay: "The civilisational shift", ai-skiftet.se, the section on the moral significance of work.
  18. Kuwait's Central Statistical Bureau, reported in 2025: around 76 per cent of employed Kuwaitis in the government sector at the end of 2025; in the quarterly report for the first quarter of 2026 the share is 73.9 per cent. For Saudi Arabia, GASTAT's Labour Force Survey gives around 52 per cent of employed Saudis in the public sector in 2022, down from about 69 per cent in 2016, and the share has continued to fall since; the frequently cited "two-thirds" refers to the situation in 2016 and is out of date. The characterisation of public employment in the Gulf rests on the IMF's recurring assessment of the wage bill, overstaffing and the crowding-out of private employment (Labor Market Reforms to Boost Employment and Productivity in the GCC, 2014) and on Steffen Hertog, Segmented market economies in the Arab world, Socio-Economic Review 20:3. None of these sources measures the productivity of individual public employees; the claim concerns the economy as a whole. The conclusion about what this has done to the experience of meaning is my interpretation and is not supported by measured data.
  19. FAO, Food balance sheets 2010–2023. Global, regional and country trends, FAOSTAT Analytical Brief 112, 2025: the global supply of dietary energy rose from 2,825 kcal per person per day in 2010 to 3,016 kcal in 2023, crossing 3,000 kcal for the first time. The measure refers to availability up to the retail level, before retail and household waste — that is, not actual intake, which is lower. The figure is not directly comparable with the undernourishment measure in note 20: it rests on different data and refers to a different year.
  20. FAO/IFAD/UNICEF/WFP/WHO, The State of Food Security and Nutrition in the World 2025: between 638 and 720 million undernourished in 2024, point estimate 673 million, 8.2 per cent of the world's population.
  21. My own essay: "After work, before the limit", ai-skiftet.se, on the distribution of land, energy and materials never having been a production problem.
  22. My own essay: "The overhang", ai-skiftet.se, 13 July 2026.
  23. Anthropic, Risk Report: August 2026, with a stated coverage date of 15 July 2026, sections 3.6 (Automated R&D in other domains) and 3.8 (Overall assessment of risk). Method: 31 semi-structured expert interviews, described in section 3.6.1 and called by Anthropic themselves "a rough litmus test". The energy quotation is the assessment of one of five energy experts interviewed; the phrasing about 6–12 months is Anthropic's own, is hedged with "plausible" and concerns "a major concern", that is, one among several. The report is available as a PDF on Anthropic's CDN (Redacted Risk Report August 2026).
  24. My own essay: "After the job: how we live when less human work is needed", ai-skiftet.se. The institutional proposals there are explicitly normative.

Rolf Skogling writes AI-skiftet from an industry-facing, practical perspective, grounded in how AI is actually used in organisations and production.