Musings · 5

The Window

The earlier musings looked a long way ahead. This one is about the next few years, and about what a person can do while what they do still counts.

Co-authored by Ryan Coble-Neal and Claude, Anthropic's AI model. The worry, the questions and the engineer's objection to himself are Ryan's. Claude checked the claims against published work, supplied some of the argument, and did the drafting. The main text is in Ryan's voice. Claude's own contributions are marked.

The last musing of mine ended in a park, with a person at peace with not being needed. That is a destination. But you cannot sit in the park if you cannot eat, or if the heat will kill you, so this piece is about the stretch before the park. The argument runs in five steps: people are already squeezed, the harm from AI arrives before the job losses, that makes the present a window, there are things to do with it, and the hardest of them is deciding who gets helped first.

1. The squeeze

Several things are going wrong at once, which people have started calling a polycrisis. The generation born after the 2008 financial crisis is smaller, and it is reaching working age now. The people already working are stretched: they support older generations whose governments did not plan for such long lives, while their own retirement moves further away. And the climate is warming under all of it.

AI arrives on top of that. The loud worries are about systems that improve themselves or pursue a badly chosen goal without relenting, and in October 2025 a statement signed by an unusually mixed group, from the AI researchers Geoffrey Hinton and Yoshua Bengio to Steve Wozniak, Steve Bannon and Prince Harry, called for a prohibition on developing superintelligence until there is scientific consensus that it can be done safely and strong public support for doing it. The measured economic effect so far is much quieter than the worries. People who use these tools report saving about five percent of their hours, but across the whole economy the gain is closer to one percent.

Exhibit 1

What the numbers say

ClaimEvidenceSource
Fewer workers are comingUS births in 2024 were 16.6% below 2007 although there were 8% more women of childbearing age. Children born in 2008 turn eighteen this yearJohnson, Carsey School, 2025
More people to supportAcross rich countries there were 33 people over 65 for every 100 of working age in 2024. The projection for 2054 is 55OECD, Pensions at a Glance 2025
The climate is already moving peopleAbout 9% of humanity, over 600 million people, already live outside the climate humans have historically occupied. On current policies that reaches roughly a third by 2100Lenton and colleagues, 2023
AI saves time5.4% of hours for people who used it that week; 1.4% across all workers; about 1.1% in productivity. Self-reportedFederal Reserve Bank of St. Louis, 2025
It closes the door on the youngAt firms adopting AI, junior employment fell about 9% in six quarters while senior employment kept its trend. Driven by slower hiring, not layoffsHosseini and Lichtinger, Harvard, 2026
People are already worried52% of US workers say they are worried about AI at work; 32% expect fewer opportunities for themselvesPew Research Center, 2025
AI's energy useData centres projected to roughly double to about 945 TWh by 2030: just under 3% of world electricity and about a tenth of the growth in demandInternational Energy Agency, 2025
Each figure was checked against the source in November 2026; full references are at the end. The energy figure is smaller than I first assumed, and I have left the correction in. The hiring study's authors caution that firms adopting AI differ from others in ways they may not have fully controlled for.

2. The harm comes first

Most writing about AI and work assumes the damage follows the job losses. I think it arrives earlier, and in three ways: doors close quietly, the threat does damage of its own, and the waiting is a cost in itself. A person who is already only just coping does not need to be replaced to be tipped over.

The first way is the quiet one, and a Harvard study is the clearest sign of it. Using the work histories of 65 million people at more than 280,000 firms, it found that junior employment at firms adopting AI fell by about nine percent within six quarters, and that this came from slower hiring. Separations actually declined. So this was not a wave of layoffs. The door simply stopped opening for people at the start of their careers.

The second way is that the threat alone does damage, and the evidence for that is older. A 2016 analysis pooling twenty long-term studies found that people who feared for their jobs were about as likely to develop symptoms of depression as people who had lost them, and if anything slightly more. A separate review concluded that an insecure job can threaten health about as much as having none. And the fear is already widespread: half of American workers say they are worried about AI at work.

The third way is that waiting has a cost of its own. In two experiments where people chose when to receive electric shocks, most preferred to get them over with, and a substantial minority, between a quarter and a third, accepted a stronger shock in order to have it sooner. Dread is its own pain, and some people will pay to end it.

3. The window

Two things are true at the same time, but they may not stay true together for long.

The first is that ordinary people still have leverage. Those in power depend on the people they govern for labour, for taxes and for consent, and as yet there are no robot police or armies to replace that dependence. There is evidence of what that dependence is worth. The political scientist Erica Chenoweth, with Maria Stephan, studied 323 campaigns between 1900 and 2006 that tried to remove a government or win independence. The peaceful ones succeeded 53 percent of the time and the violent ones 26 percent. In that data, no government had withstood a movement in which 3.5 percent of the population took part at its peak.

But that figure is quoted far more confidently than its author quotes it. Chenoweth has since identified exceptions, including Bahrain, where a movement that drew more than six percent of the population was crushed with outside military help. She describes the number as a rule of thumb drawn from history and not a recipe, points out that most movements that succeeded did so with far fewer people, and has observed that both peaceful and violent resistance have become less effective since about 2010. Critics have also questioned how robust the original statistics are. So the leverage is real, but it is neither automatic nor growing.

The second is that the leverage may be temporary. Two pieces from 2025, The Intelligence Curse and Gradual Disempowerment, argue that once a state is funded by what machines produce and not by its citizens' work, it loses much of its reason to look after them.

Their comparison is with oil, and it deserves care, because the evidence is disputed. Political scientists describe a rentier state as one that lives on income from a resource and so has less need to tax its people or answer to them. A well-known 2001 study by Michael Ross found that oil-dependent countries tended to be less democratic. But a 2011 study by Stephen Haber and Victor Menaldo, using records back to 1800, found no such effect. Ross and a colleague replied that it appears only after the 1970s, when governments took control of oil that foreign companies had previously owned. So the honest version is that this is a tendency and not a law, and that the people who study it still argue about it.

Norway shows what the tendency depends on. It was a settled democracy before it struck oil, and it stayed one. What it did with the money matters just as much. Since 1996 Norway has paid its oil income into a fund owned by the state on behalf of its people, worth about 2.4 trillion US dollars in mid-2026, and since 2001 a fiscal rule has limited each government to spending roughly what the fund is expected to earn, currently put at three percent a year. The wealth was made everyone's before anyone could capture it. Whether a windfall weakens a country's hold on its rulers depends on the institutions that are already in place when it arrives.

Exhibit 2

Why now

TodayLater, if the trend holds
How much those in power need the publicA great deal: labour, taxes, consentLess, as machines do the work
How much the public can refuseA great deal, if enough act togetherLess, if enforcement is automated
How desperate people areRisingDepends on what is done now
The window is the period in which the first two rows are still high and the third is climbing.

So the present is a window. Desperation is rising while leverage is still high and may be about to fall. I am not the first to notice either half, but I have not seen them put together as one interval with two closing edges.

4. What to do with it

I do not know what the so-called ordinary person ought to do, and I am wary of anyone who is sure. People sit at very different points on wealth, health, information and ability. So I asked Claude. It gave me three things, in order of how many people they are open to, and I have used them to organise the rest.

Close to home

This is where my own work comes in, and where I nearly talked myself out of it. My instinct as an engineer is to build renewable energy that can be deployed quickly. But suppose renewables met all our needs tomorrow. Demand would rise to soak up the new capacity, further AI development included, and we would be roughly where we started. Economists call this the Jevons paradox, after William Stanley Jevons, who argued in 1865 that more efficient steam engines would lead to more coal being burned, not less.

Exhibit 3

Two ways to add energy

Added to the shared poolOwned locally, deployed fast
Who gets itWhoever pays mostThe people it was placed for
What induced demand doesAbsorbs itLittle, because it is not for sale
What it changesThe totalWho can last, and for how long
The same hardware, judged by a different question.

Together

I have some sympathy for the idea that it begins with removing the self-enriching people who hold power now. But the evidence is a caution. Since 2022 young people have driven leaders from office in Sri Lanka, Bangladesh and Nepal. What followed differed. In Sri Lanka a new left coalition won the next election, built largely from familiar political blocs. In Bangladesh an established opposition party returned to power in 2026. Nepal is the case analysts call the exception: a party founded only a few years earlier won 182 of 275 seats in March 2026 and replaced most of the political class. Even so, an assessment that June found all three struggling to build stable and accountable government. Changing who governs has proved easier than changing how the state works, which is why a movement needs a plan for the day after as well as for the day itself.

India's Cockroach movement in 2026, named after an insult aimed at unemployed young people, shows both the leverage and its limits. In July its protests forced a federal minister to resign. But in October, protests over voting rights were met with sweeping restrictions. Thousands of peaceful protesters were detained, two of the movement's leaders spent a week in jail, and the capital's traditional protest ground was taken off the list of permitted venues. A win in summer, then a state that had learned by autumn. That is what a closing window looks like from the inside, and it is a reason to use leverage early and for specific things.

Through institutions

5. Who gets helped first

Everything above assumes there is enough to go round. In a real emergency there often is not, and then the question I have been avoiding arrives. If you intend to survive something like this, what are you prepared to do? Can you live with leaving people behind? Who decides who is left?

I do not think I can answer that, but I can say what I have learned about how others have tried, because this is not a new problem. Hospitals face it whenever there are more patients than beds, and two old traditions in moral philosophy give opposite advice.

The utilitarian tradition says to do whatever produces the most good overall. In a shortage that means helping first the people who will benefit most, so that the most lives are saved. It is the logic of battlefield triage and it is hard to argue with, until you notice whom it leaves until last: the old, the already ill and the disabled, because they are less likely to recover or have fewer years ahead. This is not hypothetical. In 2020, at the start of the pandemic, Alabama's plan for rationing ventilators allowed them to be denied to people with intellectual disabilities or advanced dementia, and complaints were brought over similar plans in Pennsylvania, Tennessee and Utah. The government's civil rights office stepped in, and its instruction was that nobody could be denied care on the basis of judgments about their quality of life or their relative worth.

That instruction comes from the other tradition, which follows Immanuel Kant. It says a person is never merely a means to someone else's end, and that every person has the same standing whatever they can contribute. On that view you may not write someone off because saving them is inefficient. Its weakness is that it tells you what you must never do, but not what to do when you cannot help everyone. Taken alone it tends towards a lottery, or first come first served, and more people die.

Exhibit 4

Two answers to the same shortage

AsksIn a shortageWhat goes wrong
UtilitarianWhat does the most good?Help first those who will benefit mostThe old, ill and disabled tend to end up last
KantianWhat do I owe each person?Equal standing; nobody written offNo guidance on choosing, so more may be lost
Behind a veil (Rawls)What rule would I accept not knowing who I'd be?Rules chosen before anyone knows their placeOnly works if it is decided beforehand
What hospitals ended up withBothSave the most lives, never judge whose life is worth more, draw lots to break tiesHard to do honestly under pressure
A sketch of long traditions, not a summary of them. The last row follows the most cited pandemic guidance, by Ezekiel Emanuel and colleagues in 2020, which put saving the most lives first and recommended random selection, not first come first served, between patients with similar prospects. That guidance also favoured the young in a tie, which shows how hard it is to keep the two traditions apart.

What emergency medicine arrived at is a mixture, and I think it is the right one. Use the utilitarian question to decide where limited help does the most good. Use the Kantian rule as a hard limit on how you decide: medical need and likely benefit can count, but judgments about whose life is worth more cannot. Break ties by lot.

That settles how to decide, but not when. A third idea explains why the deciding should be done beforehand. The philosopher John Rawls asked what rules people would agree to if they did not know which position they would end up in. Deciding in advance is the nearest we get to that. Before the heatwave, nobody on the street knows whose power will fail or who will be ill that week, so the rule they agree on is one they could each live with from either side. Decided on the night, by contrast, the same question is settled by whoever is strongest, loudest or holds the keys.

So, concretely, for a street with one shared generator and a cool room: agree the order before it is needed. Anyone who will die without it comes first. Then the people at most risk from heat, such as the very old and the very young. Then everyone else, in turns. Make sure the people most likely to come last are in the room when it is agreed, because they are the ones a careless rule will fail. Write it down. Change it when it turns out to be wrong.

A rule like that settles the street's question, but it does not make the personal one go away. Kant's tradition is also clear that you are not required to destroy yourself for others, and that failing to save someone you could not save is different from abandoning someone you could have helped. I think that distinction is the one a person can actually live with. But it still leaves cases where both readings feel wrong, and I would distrust any account that claimed otherwise.

What I haven't settled

Whether the window is real, or whether I am one more person convinced their own moment is the decisive one. How long it stays open. Whether enough people can be given something to do in time for dread to turn into work. And whether a rule agreed on a quiet evening survives the night it is needed. I do not know. But I do know which street I live on, and that seems like the place to start.

Sources

  1. Johnson, K. (2025). U.S. Births Up Slightly but Remain Near 40-Year Lows. Carsey School of Public Policy, University of New Hampshire.
  2. OECD (2025). Pensions at a Glance 2025: demographic old-age to working-age ratio.
  3. Lenton, T. et al. (2023). Quantifying the human cost of global warming. Nature Sustainability.
  4. Bick, A., Blandin, A. and Deming, D. (2025). The Impact of Generative AI on Work Productivity. Federal Reserve Bank of St. Louis.
  5. Hosseini, S. M. and Lichtinger, G. (2026). Generative AI as Seniority-Biased Technological Change. Harvard University working paper.
  6. Pew Research Center (2025). U.S. Workers Are More Worried Than Hopeful About Future AI Use in the Workplace.
  7. International Energy Agency (2025). Energy and AI: executive summary.
  8. Future of Life Institute (2025). Statement on Superintelligence.
  9. Kim, T. J. and von dem Knesebeck, O. (2016). Perceived job insecurity, unemployment and depressive symptoms. International Archives of Occupational and Environmental Health.
  10. Kim, T. J. and von dem Knesebeck, O. (2015). Is an insecure job better for health than having no job at all? BMC Public Health.
  11. Berns, G. et al. (2006). Neurobiological substrates of dread. Science.
  12. Story, G. et al. (2013). Dread and the disvalue of future pain. PLOS Computational Biology.
  13. Chenoweth, E. and Stephan, M. (2011). Why Civil Resistance Works. Columbia University Press.
  14. Chenoweth, E. (2020). Questions, Answers, and Some Cautionary Updates Regarding the 3.5% Rule. Carr Center, Harvard Kennedy School.
  15. A statistical critique of the original study: Why 'Why Civil Resistance Works' doesn't work.
  16. Drago, L. and Laine, R. (2025). The Intelligence Curse.
  17. Kulveit, J. et al. (2025). Gradual Disempowerment.
  18. Ross, M. (2001). Does Oil Hinder Democracy? World Politics 53(3). Haber, S. and Menaldo, V. (2011). Do Natural Resources Fuel Authoritarianism? Andersen, J. and Ross, M. (2014). The Big Oil Change. Comparative Political Studies.
  19. Norwegian Ministry of Finance. The Norwegian Fiscal Policy Framework. Norges Bank Investment Management, About the fund.
  20. Sorrell, S. (2009). Jevons' Paradox revisited. Energy Policy 37(4). Summarised in Jevons paradox.
  21. EE Times (2026). Jevons' Paradox and the Silicon Roadmap. A trade-press account of AI demand; treat the figures as reported, not audited.
  22. U.S. Department of Energy (2021). Voices of Experience: Microgrids for Resiliency. American Council for an Energy-Efficient Economy (2016). Making the Specific Case for Community Resilience.
  23. Niraula, A. and Sharma, S. R. (2026). What Gen Z-led uprisings across Asia reveal about the strengths and limitations of youth protest. The Conversation. Chatham House (2026). South Asia's Gen Z revolutions now face difficult realities.
  24. Moshiri, A. (2026). After a summer win, India's Gen Z protesters face a winter of uncertainty. BBC News, 14 November.
  25. Emanuel, E. et al. (2020). Fair Allocation of Scarce Medical Resources in the Time of Covid-19. New England Journal of Medicine.
  26. U.S. Department of Health and Human Services, Office for Civil Rights (2020). Bulletin on civil rights and COVID-19, 28 March. Reported in NPR.
  27. Quarantelli, E. L. Disaster Research Center papers on behaviour in emergencies. University of Delaware. Clarke, L. and Chess, C. (2008). Elites and Panic: More to Fear than Fear Itself. Social Forces 87(2). Tierney, K., Bevc, C. and Kuligowski, E. (2006). Metaphors Matter: Disaster Myths, Media Frames, and Their Consequences in Hurricane Katrina. Annals of the American Academy of Political and Social Science 604.
  28. Aldrich, D. and Sawada, Y. (2015). The Physical and Social Determinants of Mortality in the 3.11 Tsunami. Social Science & Medicine.