Once AI can do most thinking work, could your paycheck shrink to the price of the computing power that could replace you?

This explores what happens to the slice of the economy that gets paid out as wages, rather than as returns to whoever owns the machines, once AI can do most of the thinking work. The sharpest answer in the collection is that wages stop tracking how valuable a person's work is and start tracking how cheap it would be to have a computer do the same job What happens to human wages in an AGI economy?. Automation goes after bottleneck tasks first, and compute keeps getting cheaper, so labor's share of GDP heads toward zero. Some human work survives, but mostly as 'accessory' work: people keep it because spending compute on it isn't worth it, not because humans are irreplaceable. In that world, your paycheck is capped by the price of the GPU hours that could stand in for you.

The transition to that endpoint looks less dramatic and more uneven. Anthropic's own scenario modeling finds that labor's share falls and capital's share rises as AI growth speeds up Does AI growth inevitably shift wealth away from workers?. The counterintuitive part is that average wages can still rise while this happens. The pie grows fast enough that workers get a smaller slice of a bigger pie, while knowledge workers in particular stagnate or lose ground. Firm-level evidence suggests the shift won't be spread evenly. Companies that are already deep into AI replace contract workers faster and more cheaply than other firms Do firms substitute labor for AI at different rates?, so the gains pile up where AI capability already sits.

There's a near-term counterweight. When AI exposure is concentrated on a few tasks within a job instead of spread across all of them, workers move over to the tasks AI doesn't touch, and net job losses stay modest Does concentrated AI exposure enable workers to adapt and reallocate?. That cushion depends on there still being plenty of tasks AI can't do. A superintelligence economy is basically defined by that buffer running out. The benchmark literature adds a timing caveat: agents that win contests still fail at long, real-world professional workflows Why do agent benchmarks not predict real economic value?. So how fast labor's share falls depends on closing that gap, not on headline capability scores.

The idea you may not expect is that falling labor share is more than a distribution problem. It's also a loss of control. Institutions stay roughly aligned with what people want partly because they depend on human workers who care how things turn out. As AI takes over that labor, society loses that quiet steering mechanism, and systems can drift away from human preferences in ways that may not be reversible Does incremental AI replacement erode human influence over society?. A shrinking wage share therefore means a shrinking share of influence too. That's why the policy-minded notes stress choices rather than fate. Whether generative AI widens or narrows inequality depends on access, integration, and incentives, not on the technology itself Does generative AI inevitably worsen or reduce inequality?. One line of argument also holds that models built from humanity's shared written output are a collective inheritance, and fencing them off as private property creates a new kind of inequality Should restricting AI access create new kinds of inequality?. That argument opens the door to claims on AI income that don't run through wages at all.

The collection doesn't model the full superintelligence endpoint in detail. It is strongest on the wage-equals-compute-cost logic and on the early, transitional data. For the different routes AI could take to superintelligence, and the bottlenecks on each that would set the pace, see What bottlenecks define the path from AGI to superintelligence?.