Automation can make a job pay more by making it harder, or pay less by making it easier — which path do most companies choose?

This explores what happens to pay and jobs when automation removes the expert parts of a job compared with the routine parts. The corpus's central finding runs against intuition. When machines take over the routine tasks, the work that remains is the harder work. Each remaining worker needs more expertise, so wages rise, but fewer people qualify for the job. When machines take over the expert tasks, the job becomes easier to do. Wages fall, but more people can now do it Does automation raise or lower the skills that remaining work demands?. So 'automation destroys jobs' and 'automation lowers pay' are separate outcomes, and they can pull in opposite directions. Automating routine work can be good for the experts who keep their jobs and bad for everyone shut out. Automating expertise can open a job to newcomers while making it pay less.

Which kind of automation do firms actually build? Acemoglu, Autor and Johnson argue that firms have a strong incentive to automate expertise, because replacing scarce, expensive skill pays off more than inventing new tasks for workers to do. Each firm acting in its own interest adds up to an economy that underinvests in AI that would make workers more valuable Why do firms build automating AI instead of pro-worker AI?. If they're right, the wage-lowering branch of the framework above is the one the market is heading toward.

The effects also play out over time, which a snapshot of one moment misses. One model shows that automating entry-level tasks can raise output today while slowing growth later. It isn't that juniors lose their jobs. It's that novices spend less time working alongside the experts who pass on hands-on, hard-to-write-down know-how Can automation raise output while slowing growth?. Removing routine work does more than change today's wages. It can thin out the path by which people become experts in the first place. Anthropic's own engineers raise a similar worry: handing routine coding to Claude may erode the practice they need to catch its mistakes Does AI assistance erode the skills needed to oversee it?.

How concentrated the change is also matters. When AI touches only a few tasks within a job, workers can shift to the tasks that remain, and net employment barely moves. When exposure spreads across most of a job, demand for labor falls Does concentrated AI exposure enable workers to adapt and reallocate?. Firms also misjudge what they've automated. One HR-vendor survey found that most companies that made AI-driven layoffs rehired over half of those roles within six months. That suggests the AI handled simpler tasks than managers assumed and left the expert work undone Do AI layoffs actually save money for companies?. Exposure also isn't spread evenly across workers. In female-dominated occupations it reaches all skill levels, so lower-paid women are exposed at the same rate as better-paid colleagues, and they have fewer resources to adapt Does AI exposure hit low-wage workers harder in some fields?.

The takeaway you might not have expected: the useful question about any automation isn't 'how many tasks did it take?' but 'did the job left behind get harder or easier?' That one question predicts whether pay goes up or down, and whether the job opens up to more people or closes to fewer.