AI Cuts Hours Before It Rebalances Them

A couple of weeks ago, Alex Tabarrok wrote a short post, "AI, Unemployment and Work", that framed the AI-and-jobs debate cleanly: "60% of people employed and 40% unemployed is the same number of working hours as 100% employed at 60% of the hours." A few lines later he got to the real question: "the difference between catastrophe and wonderland boils down to distribution."

That is a useful correction to a lot of AI panic. People often talk as if there are only two possible futures. In one, the machines take the jobs and the economy breaks. In the other, productivity rises, everyone gets richer, and some version of Keynesian leisure arrives on schedule. The arithmetic behind those futures is closer than the rhetoric suggests. If the economy needs fewer human labor hours, there are many ways to allocate the result.

You can distribute it as mass joblessness. You can distribute it as shorter workweeks. You can distribute it as later retirement, more holidays, fewer low-value meetings, longer parental leave, and higher output with less time spent pretending to work. You can also distribute it, at least for a while, through a miserable middle state where firms take the gains and workers absorb the shock.

That last version is the one I feel closest to in actual operating conversations.

I lean optimistic on AI's long-run effect. I do think these systems will make a lot of people more productive. I do think we are underestimating how much drudge work, coordination overhead, and medium-skill procedural labor will get compressed. I do think the economy can absorb bigger changes in work than most people assume. Tabarrok is right to point out that the United States already sustained a large decline in annual working hours over the last century without settling into permanent mass unemployment. There is real historical evidence for adaptation.

The reason I cannot fully join the sunny version is that I spend a lot of time in calls with operators who are already making the first pass at the transition. What I hear there is not a plan for broad-based leisure. It is a plan for fewer people.

Recent conversations have included versions of the following: an 11-person team that could shrink to 2 as automation matures. One person handling 60 accounts where 8 used to be the rough ceiling. An 8-person team that, if AI tools were used seriously, might really be a 1- or 2-person team. No one says this in the language of social theory. They say it in the language of headcount planning, margin pressure, and what they can now get away with.

This matters because the transition path will shape the politics. A future with much less human labor can be good. A future where the gains arrive first as selective firings, frozen hiring, and surviving employees being told that AI has made a permanent speedup possible is going to feel bad even if the macro story ends up fine. People do not live inside the long run. They live inside the quarter when the team gets cut in half.

There is also a coordination problem here. A 3-day week is only partly about the amount of leisure. It is also about whether leisure arrives in a socially usable form. Tabarrok made a similar point years ago in writing about holidays and coordinated time off. Free time is more valuable when other people have it too. Scattered underemployment is not the same thing as shared leisure. A worker who loses hours because demand fell or software replaced part of the role has more free time in a technical sense. That does not mean they received the good version of leisure.

This is why I am skeptical of the glib move where people say AI will eliminate jobs and then wave toward some future abundance. Maybe. But between here and there, firms will use the tools they have for the incentives they face. Most management teams are not going to wake up, discover a 30% productivity gain, and spontaneously convert it into more holidays. They are going to tighten the org chart. They are going to push each manager to cover more. They are going to decide not to backfill. They are going to call it efficiency, and in many cases they will be right.

That does not make the optimistic case wrong. It means the optimistic case needs a theory of transition, not a theory of eventual abundance. If we want the gains from AI to show up as broad reductions in work instead of concentrated reductions in workers, that will take coordination. Some of that is policy. Some of it is labor-market design. Some of it is cultural. Some of it is firms deciding that higher productivity can buy better jobs instead of just leaner payrolls.

I do not know how this lands. That is the honest answer. I can see the case for a much better world with less drudgery and more leisure. I can also see, very clearly, the early signs of a rougher distribution where a lot of the first-order effect is headcount compression.

Maybe Tabarrok's formulation is the right way to hold both ideas at once. The reduction in labor hours may be coming either way. The real fight is over who gets the upside, who eats the transition cost, and whether reduced work arrives as freedom or as somebody getting fired.