The Exit Induction: With AGI You Don't Win the Economy — You Exit It.

Introduction: An idealized toy model

We start with a cartoonish premise. One man — call him the Owner — controls artificial general intelligence (AGI). Robotics has matured and energy is cheap and abundant. So the Owner’s machines can perform any task a human can do and the marginal cost of adding another machine-worker is minimal.

Is the Owner rich?

The naive answer is yes — he owns the most valuable asset in history after all! But this isn’t so obvious upon closer examination. Wealth, in an exchange economy, is the ability to command other people’s output. For example, stock certificates and their cash dividends can be exchanged for other goods, services, or assets. But the Owner doesn’t need anyone else’s output. He can produce whatever he wants at minimal cost, in whatever quantity he wants, including more cognitive and physical machine-workers.

What would the Owner buy? And if he buys almost nothing, in what sense does the economy still contain him?

This essay is a logical induction on a deliberately idealized model (Part 2 will add back realistic frictions). We first examine when one Owner receives AGI, and conclude that he exits the economy. Next, we consider when a second owner receives AGI, and then we run the induction until the end and everyone has acquired AGI. Economic exit is total, frictionless, and complete.

Using this idealized model we can then explore the four equilibria that are compatible with a post-AGI world, depending on whether AGI access is diffuse or concentrated, and whether the political order is stable or unstable. These four equilibria are: Wonderland, Fortress City, Neofeudalism, and Mad Max.

We then answer the standard economists’ objection of comparative advantage — “humans will always have something to do” — and show that it fails if we take the promise of AGI at face value because the value of the gains from trade falls in inverse proportion to the productivity gap between trading partners.

The next essay then restores friction to the model, showing us that the world after AGI is not a world without markets but of minimal markets, how the real world differs from the idealized model, and what policy responses are desirable and possible.

P(1): The Owner

Commentators everywhere are asking: will humans still have jobs? Will their wages suffer? This has motivated countless papers on the “Economics of Transformative AI,” productivity measures, deployment bottlenecks, and comparative advantage.¹

A simpler — and thus better — question is, why would someone with AGI remain in the economy at all?

Wealth is a claim on other people’s production. Money, equity, bonds, cryptocurrency are all IOUs redeemable against things other people make. Even in a world where only the Owner has AGI, he has no use for those IOUs because there is almost nothing other people make that his machines cannot make at a similar, if not lower, cost. He might accumulate a large fortune, denominated in claims on human output, but it would be worthless to him. Not because it is too small, but because it is too redundant.

So the Owner does not “win the economy”. He exits it.

He becomes semi-autarkic, a one-man economy trading with no one because trade offers him nothing. It turns out AGI, at the limit, makes one not quite “rich” but self-sufficient.

As he exits the economy the Owner does a final “Conversion,” which is to say, he trades away all the things he doesn’t need for things he still does. Most likely he will sell all claims in the “reproducible” part of the economy, such as equity, bonds, office towers, to acquire “irreproducible” things he needs, such as beautiful land to live on, long-term energy access, raw materials, water rights, etc. It is like trading in the casino chips for things that he or his AGI can use.

P(k) ⟹ P(k + 1): The Exit Queue

Now give a second person the same AGI fairy bundle: AGI, robotics, cheap energy. She faces the same incentives as the first Owner. And so if the first Owner exited she will also exit by doing the same things. Her claims on other people’s output will also become redundant, she withdraws from trading reproducibles (including her own labor), she converts her wealth into irreproducible goods on the way out. She exits. In fact, her exit makes it more urgent for the next person to exit. We can deduce three important things about the Exit Queue:

1. First, each exit is individually rational. Each exit is a household’s sensible response to their incentives. Markets shrink in importance one rational defection at a time. The invisible hand waves goodbye.
2. Second, the Exit Queue is ordered by wealth. AGI will not arrive everywhere at once but will glide down the price curve. This means that those at the start of the queue have the most wealth to convert into the irreproducible goods that will appreciate in the future. Many of them have already been pre-positioning for years in places like Montana, and we got a preview of this during COVID.
3. Third, each exit changes the terms of conversion for anyone still in the queue, making it more urgent the further down the queue you are. Every conversion entails selling the same things (claims on the reproducible economy) and buying the same things (land, energy, commodities). So the terms of conversion deteriorate monotonically down the queue. In other words, the later you exit, the more you are selling what everyone is selling to buy what everyone is buying. By the time AGI truly reaches the median household, the old middle-class wealth — say, a house in a suburb priced for its commute, the pension invested in consumer-facing equities — might convert at catastrophic rates.

As we can see, these three properties imply path dependence and increasing urgency of exit. The distribution of assets at equilibrium depends entirely on the actual order in which people exit — who converted early, who converted late, and who panic-sold.

It is important to note that a different “curvature” on the exit queue will still produce the same kind of world, but potentially very different final distributions. As we will see in Part 2, the Exit Queue and the Conversion trades provide powerful leverage points for policy intervention.

P(n): The Exit Induction

Once we run the induction to completion and everyone has access to AGI, what does that world look like? Again, this is an idealized model; Part 2 will explore how the real world will deviate from the ideal equilibria.

Reproducible goods price towards zero. Anything the AGI fairy can make — which is nearly everything, including more fairies — can be conjured up at minimal cost by anyone. Even productive capital becomes self-competing because machines that can build more machines destroy their own scarcity value. Software, logistics, even factories, all reprice downwards because of abundance, not despite it.

Trade will retract to irreproducible goods. Energy and commodities will be cheap but never free. Status goods remain scarce by definition, as Fred Hirsch observed half a century ago — there is only one front row and only one original.²

Measured GDP will decouple from actual welfare. In the premodern era, commercial trade was a small share of the total economy, since most output was directly consumed. After AGI, the same might be true again. There will be a vibrant market for commodities and ultra-luxuries, but most of the economy will not be visible through GDP, since it will be produced and consumed by users of AGI.

Land will re-price, but the agglomeration premium will vanish. Land is an obvious irreproducible good. But most urban land value today is agglomeration rent — proximity to jobs, to commerce, and other productive people. As people exit, these rents unwind and the value of office districts craters. In contrast, land will re-price on amenity and position; that is, coastline, climate, beauty, and proximity to the people you want near you. We already saw a miniature version of this during COVID, when remote work made it rational for the wealthy to buy and work out of properties in Hawaii, Montana, and Japan.

Four possible equilibria

The end state of this process is not unique, however. One simple way to characterize the possible equilibria compatible with the post-AGI world is to use a two-by-two matrix. On one side we have AGI access — is it diffuse or concentrated — and on the other we have political order — is it strong or weak.

Wonderland. The good equilibrium, which surprisingly enough, both the US and China are likely to reach. The Americans are following the market-heavy path, where ferocious competition among labs and distributors commoditizes AGI, making the queue relatively “flat” so that everyone gets AGI at roughly the same time and minimizes the conversion penalty.

China is following the state-heavy path, which uses tight controls to push automation and diffusion as fast as possible, much as it has already done through its robotics subsidy campaigns starting in 2014. State direction might lead to over-production and glut, but this too can be managed through “anti-involution” coordination and economic planning.

Fortress City. If AGI, especially robotics, does not diffuse broadly enough, you might end up with a small state entirely captured by ultra-productive Owners. The state provides security while the majority of citizens are outside the walls, so to speak, stranded in residual sectors. This rentier-state logic implies a sovereign funded by machine output rather than taxes, and thus a minimal need to bargain with the population. In the worst case this might look a lot like a caudillo state, in the best case it would resemble a Gulf monarchy.³

Neo-feudalism. Feudalism was not concentrated power but distributed power. Distributed AGI fairies (and weapons) without a strong central authority would reproduce exactly that structure, but at vastly higher living standards.

Mad Max. If the state dissolves and only a few Owners emerge, productive AGI becomes a prize to be seized by force rather than rightfully-held property. This is the logic of a warlord state, of 1990s Somalia.

The risks, and who ends up where

Collapse risk will concentrate at the periphery. The economies most exposed depend on selling labor abroad — remittance economies, outsourcing hubs, and late developers whose manufacturing ladder has been foreclosed by deindustrialization.⁴ They face a potential wage collapse without the fiscal capacity to bridge it.

Americans are not at risk of civil dissolution. It is popular to forecast internal disorder in the United States, but there is little evidence for this. First, the US already does far more redistribution than commentators give it credit. Federal transfers and taxes cut income inequality by roughly a quarter,⁵ transfers constitute the bulk of federal spending, and the revisionist inequality literature finds after-tax, after-transfer inequality has risen far less than alarmist headlines suggest.⁶ Second, the civil-conflict literature contradicts an American breakdown scenario. Insurgency is predicted by state weakness — low income, weak policing, rough terrain — not by grievance or the availability of small arms. This is simply not a risk for rich, high-capacity states.⁷ AGI, if anything, is a one-sided upgrade to American state coercive capacity.

The residual American risk is a possible policy lag if wages do begin to collapse. But the COVID crisis suggests that the American state can act quickly too. Roughly five trillion dollars of direct and indirect stimulus moved fast enough that poverty fell even as unemployment reached fifteen percent.

In contrast, Beijing has a weak record in direct transfers to households, preferring producer subsidies whenever possible. The Chinese growth model has relied on suppressing the household share to fund investment.⁸ Household consumption sits near 39% of GDP vs around 60% everywhere else, and the country provides only a thin safety net. The problem is that an AGI-triggered demand collapse cannot be solved by further production subsidies. But who knows, maybe the CCP will quickly change course and finally provide wealth and income transfers to households.

Economists object — comparative advantage!

The standard economist’s objection to this idealized model is comparative advantage. David Ricardo’s famous theorem states that even if one party is better at everything, such as our AGI Owner, both parties still gain from trade.

Why should this be so? Because what matters is relative opportunity cost, not absolute productivity.⁹ This theorem is constantly invoked in AI debates today. Noah Smith wrote a semi-viral post arguing that “because of comparative advantage, it’s possible that many of the jobs that humans do today will continue to be done by humans indefinitely, no matter how much better AIs are at those jobs.” And “it’s possible that humans will be paid more and more for those jobs as time goes on.”¹⁰

Tyler Cowen and Alex Tabarrok made pretty much the same points in a June 2026 interview at OpenAI.¹¹ Tabarrok argued that there is little evidence of permanent AI job destruction or automation in general. “Permanent jobs destroyed by the loom, the tractor, Excel? A big fat zero.” Trade will persist even if AI is better than humans at every possible task, he claimed, “so long as there’s some limit, like a time limit, on what they can do,” since this creates an opportunity cost between different applications of AGI. Cowen for his part went further and argued that worrying about distributional issues arising from full automation is like asking people in 1800 to manage 2026’s wealth.

Understanding Ricardian gains from trade

In his 1817 foundational work, Principles of Political Economy and Taxation, Ricardo illustrated this using England and Portugal, who only trade wine and cloth with each other, and measured their costs as the labor required to produce those two goods.

	Cloth cost	Wine cost	Total budget/year
Portugal	90 yrs/unit	80 yrs/unit	170 yrs
England	100 yrs/unit	120 yrs/unit	220 yrs

In isolation, Portugal and England each produce one unit of Cloth (or C) and one unit of Wine (or W), meaning that they spend 390 man-years to produce 2 C and 2 W. (All calculations are in the appendix)

If instead each country focuses on the good with the lower domestic opportunity cost, England produces only C, Portugal only W, and the world produces 2.2 C and 2.125 W, making everyone richer.

Who captures these gains? This depends on the “Terms of the Trade”, which is the international exchange rate between W and C. This rate is somewhere between Portugal’s domestic rate of W = ~0.888 C and England’s domestic rate of W = 1.2 C, which reflect relative domestic labor costs between these goods. At the lower bound, England captures all the gains. At the upper bound, Portugal does. And of course, we assume that each country consumes at least one unit of each good.

We can easily grasp the value of these gains by denoting them as a percentage of each country’s labor budget. If England captures all the gains from trade, it gains 35 yrs against its 220 yrs budget, making it ~15.9% wealthier. If Portugal captures all the gains, it gains 28 yrs against its 170 yrs budget, making it ~16.5% wealthier.

	1 W = 0.888 C	1 W = 1.2 C
Portugal’s Gains	Zero gains	Maximum gains: 28 yrs (16.5%)
England’s Gains	Maximum gains: 35 yrs (15.9%)	Zero gains

The vanishing value of the gains from trade

The problem with the standard economists’ argument is that the value of the gains from trade is bounded by the less productive trading partner. So as AGI owners become immensely more productive, the value of the gains from trade vanishes. It is not that there are no gains to be made, but rather that they will not be worth the effort, much as most people do not bother to pick up pennies on the street.

Consider two countries. AGI Portugal, where everyone with AGI now lives, and England, which is populated by those without AGI. Let us grant Tabarrok’s premise that there is some limit to AI productivity, and thus some opportunity cost. AGI Portugal is now 1000 times more productive, not infinitely productive. Cowen dismisses the idea that high productivity will cause a demand collapse, which is the scenario popularized by Citrini Research. “That’s a view which all economists think is wrong. It’s not really a matter of opinion.” The reason is because “there’s only huge profits at the top if there’s people at the bottom buying the product.” This circular relationship is what economists call Say’s Law.

So let’s humor Cowen and consider two scenarios then. In scenario 1, AGI Portugal consumes 1000 times more than it did before AGI. In scenario 2, AGI Portugal consumes the same bundle of goods as before, and its citizens simply enjoy their time off.

	Cloth cost	Wine cost	Total budget/year
AGI Portugal	0.09 yrs/unit	0.08 yrs/unit	170 yrs
England	100 yrs/unit	120 yrs/unit	220 yrs

In isolation and if AGI Portugal increases its consumption, AGI Portugal produces 1000 C and 1000 W while England produces 1 C and 1 W. If AGI Portugal does not increase its consumption, then each country produces 1 C and 1 W. (All the calculations in the appendix)

So what are the potential gains from trade? In scenario 1 AGI Portugal consumes at least 1000 C and 1000 W, so its maximum gains from trade are 0.2C and 0.125 W. In scenario 2 AGI Portugal consumes at least 1 C and 1 W, and its maximum gains from trade are still 0.2 C and 0.125 W.

	1 W = 0.888 C	1 W = 1.2 C
AGI Portugal’s Gains	Zero gains	Maximum gains: 0.028 yrs (0.0164%)
England’s Gains	Maximum gains: 35 yrs (15.9%)	Zero gains

Now we can see the wrinkle. Whether the AGI Portuguese choose leisure or debaucherous consumption, the value of the gains from trade is bounded by their less productive trade partner.

This is counterintuitive, but consider that everything that England produces in a year would cost AGI Portugal 0.17 man-years to make. This means that even the prize for conquering England and enslaving her citizens is worth just two months of one Portuguese worker’s time. It is like asking if it is worth it for the U.S. military to conquer an isolated Amazonian tribe to confiscate their baskets.

The corollary is that if there is an explosion of productivity in one trading partner, the value of the gains from trade fall in inverse proportion to the productivity differential between the trading partners. The gains for the entire country of AGI Portugal are 0.028 yrs, or 0.0164% of its labor budget. If people still work 2,000 hrs/year, that is 19.8 minutes per year, or about 23 seconds per week. If citizens spend 90% of their time on leisure, then they would only save 2.3 seconds per week!

Quite literally, the gains of trade are not worth their time. If AGI Portugal continues trading with England, it is out of charity, not self-interest.

Conclusion: Is exiting actually easy?

The Exit Induction is an idealized model that, while cartoonish, exposes powerful incentives at work. With sufficiently cheap energy and commodities, anyone with cognitive and physical AGI can produce whatever they want. They will have no use for claims on the output of others and as a result will withdraw from most sectors of the market. The exchange economy thus shrinks through a billion individually rational exits. The extent to which these forces operate depends on 1) actual energy and commodity abundance, 2) the frictions around market exit, and 3) the terms of the exit conversion.

The economists object that comparative advantage means that there will always be gains from trade. But they ignore that the value of those gains falls in inverse proportion to the productivity gap between trading partners.

Ironically, while Ricardo’s Theory of Comparative Advantage shrinks into insignificance, his Theory of Rent, which demonstrated that economic surplus accrues to fixed-supply production factors, inherits the earth. Thus, if machines can make almost everything, including other machines, a disproportionate share of that economic surplus will accrue to things which cannot be replicated — land, energy, commodities, and, perhaps, status.

Coda: A Wage Parenthesis?

The world of the Exit Induction, though strange to modern eyes, is familiar to the historian. An economy of autarks consuming what they produce and trading only in the most expensive luxuries, where property is primarily acquired through birth, marriage, or force and where status is chained to land and position — that is not science fiction. It is most of recorded history. The transition to modernity in most of the world required powerful forces to break this older order, including coercing self-sufficient peasants to produce for the market and transforming them into consumers of fertilizers, manufactures, and credit.

That strange world will again be our own. The roughly two centuries in which ordinary labor could convert into meaningful property are an anomaly in historical time. Call it the wage parenthesis. The Exit Induction thus suggests that AGI closes this anomalous era and returns us to the historical norm, not necessarily by immiserating anyone, but by making redundant the markets which give wages their importance. Already we can see this process in places such as the semiconductor powerhouse of Taiwan, where a married couple needs to save for 40 to 60 years to buy a median apartment. Exploring the parenthesis, and what pre-wage economies teach us about the post-AGI world, is a separate essay’s work. Here it is enough to say: we have the prior knowledge.

Citation

Rincon-Cruz, Manny. “The Exit Induction.” Prior Knowledge, July 2026. https://prior-knowledge.ai/essays/the-exit-induction/.

Download the PDF version of this essay.

Appendix

Ricardo’s Comparative Advantage:

	Cloth production	Wine production	Total labor cost
Portugal	90 yrs×(1 unit/90 yrs) = 1 unit	80 yrs×(1 unit/80 yrs) = 1 unit	170 yrs
England	100 yrs×(1 unit/100 yrs) = 1 unit	120 yrs×(1 unit/120 yrs) = 1 unit	220 yrs
Combined production	2 units	2 units	390 yrs

	Cloth production	Wine production	Total labor cost
Portugal	0 yrs×(1 unit/90 yrs) = 0 units	170 yrs×(1 unit/80 yrs) = 2.125 units	170 yrs
England	220 yrs×(1 unit/100 yrs) = 2.2 units	0 yrs×(1 unit/120 yrs) = 0 units	220 yrs
Combined production	2.2 units	2.125 units	390 yrs

	1 W = 0.888 C	1 W = 1.2 C
Portugal’s Gains	Zero gains. Consumes: 1.0 C, 1.0 W (Trades 1.125 Wine for 1.0 Cloth)	Maximum gains: Consumes: 1.2 C, 1.125 W (Gets 1.2 Cloth for 1.0 Wine)
England	Maximum gains. Consumes: 1.2 C, 1.125 W (Trades 1.0 Cloth for 1.125 Wine)	Zero gains. Consumes: 1.0 C, 1.0 W (Trades 1.2 Cloth for 1.0 Wine)

AGI Portugal and England:

	Cloth production	Wine production	Total labor cost
AGI Portugal	1000	1000	170 yrs / 170 yrs
England	1	1	220 yrs
Combined production	1001 units	1001 units	390 yrs

	Cloth production	Wine production	Total labor cost
AGI Portugal	1	1	0.17 yrs / 170 yrs
England	1	1	220 yrs
Combined production	2 units	2 units	390 yrs

	Cloth production	Wine production	Total labor cost
AGI Portugal	999	1001.125	170 yrs
England	2.2	0	220 yrs
Combined production	1001.2 units	1001.125 units	390 yrs

	Cloth production	Wine production	Total labor cost
AGI Portugal	0	2.125	0.17 yrs / 170 yrs
England	2.2	0	220 yrs
Combined production	2.2 units	2.125 units	390 yrs

	1 W = 0.888 C	1 W = 1.2 C
AGI Portugal’s Gains	Zero gains. Consumes: 1000.0 C, 1000.0 W (Trades 1.125 Wine for 1.0 Cloth)	Maximum gains: Consumes: 1000.2 C, 1000.125 W (Gets 1.2 Cloth for 1.0 Wine)
England	Maximum gains. Consumes: 1.2 C, 1.125 W (Trades 1.0 Cloth for 1.125 Wine)	Zero gains. Consumes: 1.0 C, 1.0 W (Trades 1.2 Cloth for 1.0 Wine)

	1 W = 0.888 C	1 W = 1.2 C
AGI Portugal’s Gains	Zero gains. Consumes: 1.0 C, 1.0 W (Trades 1.125 Wine for 1.0 Cloth)	Maximum gains: Consumes: 1.2 C, 1.125 W (Gets 1.2 Cloth for 1.0 Wine)
England	Maximum gains. Consumes: 1.2 C, 1.125 W (Trades 1.0 Cloth for 1.125 Wine)	Zero gains. Consumes: 1.0 C, 1.0 W (Trades 1.2 Cloth for 1.0 Wine)