When videogames simulate supply chains, they usually have a model where you have discrete natural resources, which you can move through some chain of processing downstream in order to get whatever end product you want. So your video game steel mill might need access to an iron mine, a coal mine, and some laborers, and having that steel mill means you can produce cars or tanks or forks or whatever final goods are relevant to the game. This is a much simpler model than what happens in the real world, which is:

• Many metals show up in deposits together, so you're getting a mix of different materials.¹ Oil and gas are the most intuitive example; sometimes you get both, and have to figure out what to do with one or the other (before pipeline infrastructure was set up, natural gas was often viewed as a waste product and burned).

• There's a supply chain for extraction, with both durables (e.g. drills) and consumables (frac fluid, sulfuric acid). And, of course, there's the energy input required to actually get people to the production site. All of these supply chains overlap; you need a lot of steel to have an energy industry and you need a lot of energy to have a steel industry.

• The consumer products people buy are themselves often complements to labor, some of which are durable goods that pay off over time. A washing machine represents the capitalized value of the money and possibly time saving of not regularly going to a laundromat.

All of this is dizzyingly complex, especially once you start to consider the role some of the most complex supply chains—commercial aircraft manufacturing and semiconductor fabrication, for example—play in expanding the output of all the rest. Building geological models to identify promising deposits and figure out the optimal way to extract them isn't the main use of GPUs, but we'd probably have a much harder time finding more of these resources if we didn't have that full supply chain available. And semiconductor fabrication is a very high fixed-cost, comparatively low marginal cost proposition—so its size is a function of the size of the global economy.

But what all this complexity also means is that a modern economy produces lots of interesting tradeoffs. When you have limited factors of production—land, labor, manure, hand-crafted durables that are mostly made with local materials—your most interesting choice is between working and starving. You might have room to improve efficiency, and indeed pre-modern agriculture is incredibly efficient given its technological limitations, because it was developed in societies that were close to the Malthusian limit and the ones that didn't figure out how to make the best of their limited resources died out. But as soon as society reaches escape velocity, and can start producing wealth faster than population growth consumes it, there's a proliferation of choices.

It's interesting to note that one of the first export-oriented, large-scale manufacturing businesses was textiles, which is also a business where there's a lot of potential for customization. Globalization in some form is earlier than that, because there were natural resources that existed in some places but had demand everywhere, whether pepper or tin. And there were luxury products that could get bought, gifted, bartered for, or looted. But textiles were an early case where there was scaled production in some category that was accessible to people other than the very rich or very commercial—and, in fairly short order, there were many kinds of fabrics, patterns, attire, etc. So that early glimpse of a modern trade system was also an early look at how different kinds of products could be substitutes, and different kinds of labor and capital could also be swapped for one another.

That manufacturing introduced another kind of potential substitution, in labor markets: you could move from a farm to a city, switch from agriculture to manufacturing, and maybe have a better life (or at least an interestingly worse one—until the early 20th century, city-dwellers had reliably shorter lifespans). As manufacturing got more complex, and as markets developed more, there was also fungibility over time: you could jump right into unskilled work, or take a standard-of-living penalty for a while in the form of an apprenticeship or other training before opting into higher-wage work.

One way to describe the second industrial revolution is that it was a surge in the substitutability between physical capital and human labor. Mechanically-powered factories, coal-powered ships and railroads, more affordable mass-produced tools—all of these crystallize some number of saved labor-hours in a physical object. If you worked at, say, a sewing machine factory, you were basically converting your assembly-hours into a future stream of sewing-hours-saved, though both the consumers and the capitalists got the better end of that bargain than workers, for a while. And the years immediately after that, the 20s, a substantial form of the labor-saving showed up in household labor, through the introduction of appliances.

And now we're at a still more advanced stage of substitution, where you can exchange human mental labor for mechanical mental labor. As with every other stage, there are tradeoffs: the mechanized sort can beat humans at all sorts of pre-designed evaluations, but we don't yet have agents that can reliably do all of the tasks that a single human worker would have been assigned. We do have agents, APIs, and chatbots that can replace or augment substantial fractions of that work. And as with other cases of substitution, the next step, and the longest, is to rework more parts of the economy to make these substitutions optimally.

We've written about substitutability in many Diff contexts, including:

• We have pieces on the challenging economics of durable goods and why electrification took so long.

• Thoughts on what the real ROI on AI has already been ($).

• The complementarity between writing and thinking, and how LLMs are useful for producing known outputs but tricky for exploring new ideas.

• And another case where two different outputs are tied to the same input: writing code without AI might be less financially rewarding but still more fun ($).

Join the discussion!

1  This is one reason metals prices can be so volatile. Sometimes a spike in demand for X will lead to a glut of Y, because they show up in the same deposit and selling off the extra Y is still worth doing even if setting up a new Y-focused mine isn't worth doing on its own.