Venkatesh Rao argues that AI anxiety partly treats intelligence like bullion: a scarce asset to accumulate, protect and keep under elite control. In “Cognitive Mercantilism,” published on October 25, 2026, he offers a different frame. Making cognitive capability abundant could open up activities that today seem too expensive or difficult to attempt.
A Hacker News discussion of the essay exposes the harder question inside that argument: even if AI enables more production, what gives ordinary people a claim on the proceeds?
The useful distinction is between three propositions: cognitive work becomes cheaper, demand for it expands, and workers capture some of the gains. Rao supplies a reason to take the second possibility seriously. His critics challenge the leap from greater activity to broadly shared prosperity. These are separate economic questions, and treating them separately makes the debate much clearer.
Rao’s Four Scarcity Assumptions
Rao builds his analogy from his reading of Joyce Appleby’s The Relentless Revolution: A History of Capitalism (2010). In his account, the emergence of capitalism required a break with a worldview that understood wealth primarily through control of finite resources, territory and labor.
He maps that worldview onto contemporary thinking about intelligence:
| Assumption in Rao’s account of mercantilism | Its proposed counterpart in AI debates |
|---|---|
| Wealth is finite, so gaining it means capturing a larger share. | Intelligence is a scarce resource contained in brains and their products. |
| Ordinary people require supervision by governing elites. | Cognitive elites should control access because others may misuse intelligence. |
| Economic strategy centers on controlling supply and commanding high prices. | Talent, intellectual property and AI output are resources to accumulate and protect. |
| Most people have a fixed, limited productive role. | Automating routine cognitive work leaves people with little useful work to do. |
These are Rao’s interpretive categories, not a validated model of either economic history or the AI industry. His central criticism is captured in one sentence:
“The cognitive mercantilist wants to hoard intelligence and its products like bullion.”
Here, intelligence functions as an economic shorthand for capabilities used to produce knowledge, solve problems and perform cognitive work. Rao’s argument concerns how people value and distribute those capabilities. It is not a claim that every form of human intelligence can be reduced to an interchangeable commodity. His essay uses the comparison to challenge assumptions about scarcity and ordinary people’s potential.
I find the demand-side criticism more useful than the label itself. Calling a position mercantilist can make disagreement sound like historical backwardness. Asking which worthwhile activities become possible when cognitive assistance gets cheaper produces a more concrete discussion.
Cheaper Cognitive Work Could Unlock More Projects
Rao uses agricultural productivity to challenge the idea that automating an occupation makes its former workers economically useless. In his telling, improvements in farming allowed a smaller share of the population to supply food while others pursued different activities.
The relevant mechanism is that lower production costs can change what people choose to do.
Consider a hypothetical small organization that cannot justify commissioning custom software. If AI assistance makes the project affordable, the organization might build something it previously went without. That would be additional activity, not simply an existing software contract completed at a lower price.
Hacker News commenter akanet makes this defense of Rao’s argument: access to intelligence could bring more projects within reach and lead people to increase their efforts. That is an induced-demand argument. Lower cost changes the amount of a service people want to use. The exchange appears in the essay’s discussion thread.
A project’s relevant cost would include more than generating an answer or writing code. If the output requires checking, integration or ongoing maintenance, those requirements affect whether the project becomes worthwhile. This is a practical condition of the hypothetical, not a measured result reported by either source.
The demand argument therefore has real analytical value without settling the employment question. More people might commission software, pursue research or attempt other projects. The amount of paid human work involved depends on how those projects are produced.
Rao’s agricultural comparison gives readers a reason to question fixed-demand assumptions. It offers no numerical estimate of how much new demand AI will create, how quickly it will emerge, or which workers will benefit.
Ownership Changes the Meaning of Abundance
Hacker News commenter samrus challenges the analogy at the point where productive opportunity becomes economic leverage. Their argument is that earlier productivity gains left people able to produce other valuable things. Broad automation of intellectual work, potentially followed by skilled physical work, could narrow that route.
Samrus writes:
“You need to produce things to be able to reap the fruits of economic progress.”
The commenter then asks what nonowners would produce if AI owners could supply the relevant goods and services themselves. Another participant, dot42, responds by arguing against private ownership of such a powerful resource. These are individual positions in the Hacker News thread, not representative findings about technical workers or the wider public.
The strongest part of samrus’s objection is that it can coexist with abundant output. AI services could become more plentiful while particular workers lose negotiating power. A business might produce more, customers might pay less, and employees might still receive a smaller share of the resulting value.
That objection concerns ownership and bargaining power, not merely whether AI makes production cheaper.
I would give this objection more weight than a blanket assertion that new jobs cannot appear. It identifies a specific mechanism: if owners can readily replace someone’s contribution, that person may have less leverage over pay and conditions. Whether new tasks emerge is only part of the question; whether people retain a valuable role in performing them matters too.
The same distinction limits the implication that human wants will automatically support paid employment. People can want more services without possessing the income to buy them. Their purchasing power needs some basis, whether wages, ownership income or transfers. Naming those possible routes does not show which will exist at sufficient scale.
Three Claims That Need Separate Evidence
Rao and his critics become easier to compare once their arguments are separated into three steps.
Cognitive production gets cheaper
This proposition concerns the resources needed to complete a useful task at an acceptable quality level. A lower price for generated output is relevant, but the economic result also depends on how much work remains around that output.
Neither the essay nor its discussion supplies a general measurement establishing this outcome across cognitive occupations. For this debate, cheaper cognitive production is the premise being explored.
Frequently Asked Questions
3 questions
1What Does “Cognitive Mercantilism” Mean?
It is Rao’s term for treating intelligence as a scarce resource to hoard and control, while giving too little attention to the demand broader access might create.
