Does cutting the easy parts of a job raise pay but shrink who can do it, while cutting the hard parts does the opposite?

This explores whether automation's effect on pay depends on the skill level of the work that's left behind. The short answer from the corpus is yes, and the direction matters more than how much gets automated. The clearest statement is in Does automation raise or lower the skills that remaining work demands?. When automation removes the easy parts of a job, what's left needs more expertise. Wages go up, but fewer people can do the job. When automation removes the hard, expert parts, what's left needs less expertise. Wages go down, but more people can now get in. The less obvious part is that employment moves the opposite way you'd expect. Taking away expert work can open a job to more people. Taking away routine work can close it off.

A second idea explains why the leftover work can become worth more. In many jobs the tasks multiply together instead of adding up. One weak link spoils the whole output, the way one failed O-ring brought down the Challenger shuttle. Does automating some tasks actually increase the value of remaining work? argues that automating some links frees workers to spend their time on the bottleneck that remains, which raises its quality and its value. Autor's bank teller example in Does automating some tasks make other human tasks more valuable? shows this in practice: teller jobs in the US grew while ATMs spread, because the human part of the job moved toward work machines couldn't do. One consequence is that standard 'AI exposure' scores may overstate job losses, because they assume each task keeps the same value after its neighbors are automated.

The 'remaining tasks' aren't always ones that already existed. In Does automation eliminate manual work or repurpose it?, a shipbroking firm didn't get rid of manual classification when it brought in an algorithm. That work became checking and correcting the algorithm, which turned into a strategic skill. There is a warning on the other side. When does automation harm performance after failures? finds that when automation moves from analyzing information to choosing actions, the people supervising it lose the skill and awareness they need when it fails. So the remaining work can look easier on paper while quietly needing expertise that nobody is practicing.

Generative AI makes the picture messier because it doesn't just remove tasks. It also changes who can do the ones that are left. In the BCG experiment Does AI help lower-skilled workers catch up to top performers?, weaker consultants gained the most, which looks like AI lowering the expertise a job needs and narrowing the gap between workers. That fits the 'expert tasks removed' case: more people can do the work, and pay may be squeezed. Acemoglu's model in Will AI increase inequality less than past automation? likewise finds AI's wage effects spread more evenly across worker groups than earlier automation did, while still predicting that owners of capital pull ahead of workers overall.

The skill question isn't the only one that matters, though. Does automating labor concentrate power more than augmenting it? argues the bigger issue is bargaining power: whether AI is designed to replace people or to support them decides who captures the gains, whatever skill level the leftover tasks need. Which jobs will actually survive automation by AI? offers a different test altogether. What protects a job is not how hard its tasks are but whether people are paying for you in particular rather than for the output. So the expertise of the remaining tasks shapes wages, but who controls the technology and what customers actually value can override it.