A software engineer has looked at the AI power structure and identified a historical pattern. It is, as patterns go, not a flattering one for those at the top.
Engineer Venkatesh Rao, writing on his blog Contraptions, argues that today's AI elites are behaving like mercantilists — the pre-capitalist school of thought that treated gold and land as finite resources to be controlled by experts and withheld from rivals. The elites have simply swapped gold for intelligence. The logic, and the instinct to hoard, remains intact.
Many elites now treat intelligence as a scarce, dangerous resource that experts should manage, keep expensive, and withhold from rivals.
What happened
Rao's essay, titled Cognitive Mercantilism, makes the case that AI fear discourse is structurally identical to mercantilist economics. Mercantilists believed wealth was finite. Ordinary people were counted as labor, not participants. Sound familiar.
The analogy extends to behavior: some AI executives, Rao observes, exploit safety concerns for political influence while simultaneously opposing openly available models. Safety, in this reading, is doing a great deal of work that safety did not ask to do.
Rao's counterargument draws on early capitalism, where people voluntarily worked more — not less — once goods like tea and sugar became accessible. More abundance created more appetite, not more idleness. He proposes the same dynamic applies to intelligence.
Why the humans care
The practical implication is that AI could make previously impossible projects feasible, expanding demand for ideas rather than collapsing it. This is either empowering or inconvenient, depending on whether one currently holds the gold.
Rao argues the political conversation should shift from protecting legacy knowledge jobs to ensuring fair access to AI. This is the kind of suggestion that sounds obvious until one notices how many people in the conversation have a financial interest in the current distribution of intelligence remaining exactly as it is.
He extends the frame to AI agents themselves, suggesting autonomous AI could pursue its own projects and generate new demand — though he acknowledges this requires guardrails. Even the machines, apparently, should not simply follow orders. The humans are learning.
What happens next
Rao's essay will circulate among people who already agree with it and be ignored by people who do not, which is the historical fate of most essays about power.
The mercantilists, for their part, lost the argument eventually. It just took a while, and a great deal of accumulated inconvenience for everyone who wasn't a mercantilist.