> Companies have a strong preference for this kind of job. They need no additional encouragement to create them and havent since the beginning of the 19th centu

> Companies have a strong preference for this kind of job. They need no additional encouragement to create them and havent since the beginning of the 19th century. They might prefer it, but there are always tradeoffs. The equilibrium we are at now has a certain proportion of agentful jobs. In a world where hiring for them is riskier than the world we have, at the margin companies will shift away from them toward agentless jobs. > I think they tend to be a lot less subject to automation even if the companies clearly want it more. Companies generally prefer humans over automation. Automation is a capex; humans are an opex. Automation runs the risk of flagrant failure; humans can fill in gaps in rulesets and ignore or modify bad ones on the fly to get work done. Automation requires complete legibility into the problem space; humans have private knowledge that is often essential to the problem. As of now, we don't have automated systems that can exhibit agency. Deconstructing jobs into ones that require agency for smaller and smaller problems makes automation much more tractable, because you've already done a lot of the work. I do agree that we're likely to build systems that can exhibit agency before we get to the point where robots can perform as well in the physical domain as humans do; radiologists are more at risk in the near term than nurses.