Automating the routine parts of a job can leave fewer people doing it, yet those who stay often earn more — why?

This explores why automating routine work can shrink a job's headcount while raising the pay of the people who stay. The corpus's clearest answer is that wages and jobs aren't really responding to automation in general. They respond to what the remaining work demands. If you remove the easy, inexpert parts of a job, what's left is harder on average. The job now needs more expertise, so it pays more, but fewer people can do it. Both effects come from the same change. The model in Does automation raise or lower the skills that remaining work demands? also runs this in reverse: automating the expert tasks lowers the bar, pulls wages down and lets less-skilled workers in. Employment can move opposite to what a simple 'tasks lost' count would predict.

The surprising part is where those routine tasks used to sit: at the bottom of the career ladder. The easy tasks automation removes are often the ones novices learned on. Payroll data in Is generative AI displacing workers at economy-wide scale? finds no economy-wide job losses, but young workers in AI-exposed occupations are hired at rates 19% lower than their peers, while experienced workers show no such gap. That is the 'higher bar, smaller pool' pattern showing up first at the entry point. An overlapping-generations model in Can automation raise output while slowing growth? points to a longer-run cost. When novices get pulled away from the experts who pass on tacit know-how, output can rise today while long-run growth and welfare fall, even when junior employment holds steady. Higher wages for incumbents can sit alongside a weaker pipeline for producing the next generation of experts.

Whether this squeeze turns into real job loss depends on how automation is spread across a job's tasks. Firm-level data in Does concentrated AI exposure enable workers to adapt and reallocate? shows that when AI hits only a few tasks within a firm, workers shift to the tasks it doesn't touch and the net employment effect stays modest. Broad exposure across many tasks is what cuts labor demand. A related finding, When does AI actually boost worker productivity?, is that AI gains show up when workers apply skills they already have, not when they use AI to learn new ones. That favors incumbents over newcomers in the same way.

One caution: today's AI may not follow the classic 'routine task' script at all. Where workers have actually handed tasks to AI, the delegation concentrates in information-intensive work and tracks what the technology can do. It departs from routine-task automation predictions, and the wage pattern reverses at advanced degree levels Where have workers actually delegated tasks to AI?. Danish records suggest timing matters too. Two years after ChatGPT's launch, workers had taken on new AI-related tasks, but earnings and hours were stable within a 2% margin Does AI chatbot adoption change worker pay and hours?. Tasks get reorganized first, and pay and headcount move later, if they move at all. So the useful question for any given job isn't 'will it be automated?' It's 'which tasks go, and do they make the remaining work harder or easier?'