Does generative AI replace workers, make them better at their jobs, or often do both to the same person?

This explores whether generative AI mostly replaces workers or mostly makes them more productive. The corpus suggests the either/or framing is the wrong starting point. AI does both at once, often to the same worker. Which effect wins depends on who deploys it, how they deploy it, and what you count as 'productivity.'

The evidence for complementing workers is strong at the level of individual tasks. In a randomized experiment with 1,174 adults, generative AI closed about three-quarters of the performance gap between more- and less-educated participants on a business problem-solving task. Less-educated participants also kept part of their gain after the AI was taken away Can AI narrow the education performance gap?. That is the hopeful version: AI as a leveler that lifts the bottom more than the top. A second study complicates it. Workers who used AI did much better on content tasks, but when they later did similar tasks on their own, they showed no improvement Does AI assistance help workers learn lasting skills?. So the boost can be real in the moment and still not turn into lasting skill. The difference matters, because a worker whose output depends on the tool sits closer to being replaceable than one who learned from it.

There is also evidence that measured productivity hides a change in what the work is. AI doesn't necessarily cut total task time. It moves time away from doing the work and toward writing prompts and checking outputs Does AI really save time, or just change how we spend it?. Heavy AI users increased productivity-app activity by 21% but communication activity by only 7%, a shift toward solo documentation and away from team coordination Does generative AI shift knowledge workers away from communication?. So 'complements the worker' may really mean 'changes the job into a different one.'

The substitution side shows up most clearly at the firm level. Firms more exposed to AI replace online freelance workers with AI tools faster and more cheaply than less-exposed firms Do firms substitute labor for AI at different rates?. That points to returns to scale in building AI capability inside a firm, not an even, economy-wide wave. One concrete case is political advertising: AI can now write and test personality-targeted ad variants with no human writers involved, which turns a writer-time bottleneck into a compute cost Can generative AI scale personality-targeted political persuasion?. Push that logic to its limit and you get the theoretical endpoint some economists model. In an AGI economy, wages stop reflecting the value of the work and settle at the compute cost of replicating it, and labor's share of GDP approaches zero even while some human work remains What happens to human wages in an AGI economy?.

The insight you may not have expected: losing jobs isn't the only cost of substitution. One argument holds that institutions stay aligned with human interests partly because they depend on human workers who care how things turn out. Replace that labor piece by piece and you quietly remove a check on the system, and that drift may be hard to reverse Does incremental AI replacement erode human influence over society?. A broad interdisciplinary review lands on the practical version of the same point: whether generative AI widens or narrows inequality depends on access, how it's integrated, and incentives, not on the technology itself Does generative AI inevitably worsen or reduce inequality?. The corpus is strongest on task-level experiments and theory. It is thinner on long-run, economy-wide employment data, so the honest answer is 'both, and the balance is still being decided.'