In a services-oriented economy, it might be reasonable to think about cost disease as the dual to automation. Automation makes the individuals more productive at Task A, so their output gets cheaper while wages go up (at the margins – obviously, sometimes mechanical assistance doesn't do this, but many times people end up with different-ish jobs with more productivity). So why would people do Task B? Because they're getting paid a similar (training-adjusted) amount. But that higher cost per unit of output raises the incentive for automation (in proportion to the market for Task B). In a perfectly idealized market, we might expect the margin between these tasks to be in equilibrium – all tasks receiving N training-adjusted-service-dollars/year will have automation advanced to where the next marginal 1% of savings would cost the same (risk-adjusted) R&D investment.

Admittedly, that's a bit of a mouthful. And the idealization might not be the most interesting part – take, for instance, law, where there seems to be a glut of talent, reasonably high costs, and pretty marginal automation. I suspect the issue isn't productivity so much as how opaque quality is and how important the appearance of excellent advocacy can be. At the margin, to get better legal care, we might want a society with lawyers that appear more mechanistic and less like advocates – but of course, almost all actual lawyers that look that way will be (marginally) worse.

The interesting part I've been thinking about is: I don't hear much about the necessary automation-demand flux, despite being a software engineer myself. I don't know why – plenty of firms have extremely non-linear benefits from automation, and being responsive to both market size (so cost advantages are compounded across a greater number of interactions) and market wages is classic managerial work. But because businesses have non-linear automation-payoffs (the first 20% productivity improvement could be relatively cheap, and the 1% after it could be hideously expensive), we'd expect a natural flux in what needs to be done – because once the automation task is done once it's (basically) done.

Consider, what happens when we get decent automated driving? Sure, many engineers will keep developing improvements. But that also massively changes the cost of labor in many other places. Surely, some other industry will become the biggest opportunity and engineers will be lured there by well-capitalized businesses seeking to reap the rewards.

Basically, I'm saying it isn't tech's fault that it's so driven by trends, and those trends will probably end up saying more about our economy than many other models. At the end of the day, coal miners are talked about in public discourse constantly, and supply chain managers aren't (despite, I suspect, almost all economists agreeing the latter influenced the last 100 years substantially more). So a better lens might be needed to think about the topics we end up talking about – even if, on the scale of the US economy, coal miners aren't really any more important than flip-flop retailers. Because their story is the story of this innovation flux and how it impacts the economy.