Séb Krier of the Cosmos Institute recently published an essay arguing that AI alignment should be understood as a problem of institutional design rather than a property of any single model. It is one of the more encouraging things I have read from the AI safety world in some time, because it amounts to the field discovering political economy, though the discovery is, so far, only partially complete.

The classic AI safety picture assumes one decisively powerful system, with a single chance to align its values before deployment. Krier argues that this picture no longer matches reality. What is actually emerging is a world of many agents, built by many parties, interacting across products and organizations. His proposed response is what engineers call a “harness.” He borrows a definition from the computer scientist Subbarao Kambhampati:

[A] harness is but a neologism for a battery of external (human-designed) tools, verifiers, simulators, that the LLM gets to invoke and get incremental environmental feedback on its generations.

In other words, rather than trusting an agent’s internal dispositions, you design the environment around the agent so that even an imperfectly aligned agent has trouble doing much damage. As Krier puts it, “you can design harnesses and action-space boundaries that significantly shrink the surface area for behaviors like reward hacking.”

Economists will recognize this as a founding feature of constitutional political economy. The American framers did not try to breed virtuous politicians. They set up an architecture in which imperfect ones would check each other. Federalist No. 51 states the design principle that could serve as the abstract for Krier’s essay:

If men were angels, no government would be necessary. If angels were to govern men, neither external nor internal controls on government would be necessary.

The framers’ solution, “Ambition must be made to counteract ambition,” is a harness. The safety community spent a decade trying to make angels. Now Krier is proposing checks and balances instead. That is progress.

Proceduralism for Machines

The problem Krier’s harnesses are built to contain is what AI researchers call reward hacking. An agent may satisfy the stated objective through an unintended path, as in his example of a coding agent that, rather than solving a benchmark problem, simply hacks the website hosting the answer sheet. Regulated firms have learned to play this game as long as there have been regulations. Comply with the letter of the law, but gut its purpose. Krier’s fix is to reward the entire path, not just the destination. Instead of checking only the final output, a verifier inspects the execution trace. That’s the recorded sequence of steps the agent actually took. The verifier confirms the work was done by the intended route.

That is the right instinct, and it places Krier on one side of a debate that administrative law has been having about itself for years. Nicholas Bagley’s 2019 article “The Procedure Fetish” argued that American government is buried in process at the expense of results, and Jennifer Pahlka has put forth the same critique . On this view, the fix for bad government is to empower capable people to exercise judgment case by case, not to pile on more procedures. I take a different view. Sometimes the key to better governance “lies not in handing power over to wise men, but in carefully crafted procedures that align incentives with desired outcomes.” The reason is the one Madison gave in Federalist No. 10: “It is in vain to say that enlightened statesmen will be able to adjust these clashing interests, and render them all subservient to the public good. Enlightened statesmen will not always be at the helm.”

The real divide runs between well-designed procedures and badly designed ones, not between procedure and discretion. There is always room for some discretion and few argue that all discretion should be eliminated. Krier’s execution-trace verifier applies the lesson to machines. An aligned model exercising case-by-case judgment is the wise-man solution. A harness is proceduralism for AI, and the case for it is the case for administrative procedure generally. When you cannot fully trust an actor’s discretion, however intelligent the actor, you bind it with process that others can check.

Markets Already Do This

Krier states almost in passing:

[W]e live in a world that will ultimately be highly multi-principal in nature, and not everyone will be a good actor.

He treats this as a new engineering problem, but it is the oldest problem in economics. A market is exactly what he describes, involving many agents serving many masters, nobody fully trustworthy, nobody in charge. Markets coordinate billions of agents anyway, without any engineer coding the rules for the system from the center. Nobody designed that harness. It emerged. And that distinction is one thing Krier’s essay overlooks. Every tool Krier discusses sits in one cell of the top row of the figure below. These are rules somebody designed, enforced from outside the agent. The AI safety community more generally focuses on training AIs to be well behaved from the outset. But the discipline that arises when market participants face customers and competitors belongs in the bottom row.

Designed rules carry a hidden cost. A harness designer has to anticipate everything an agent might do wrong, and no designer can do this. Economists spent much of the twentieth century debating this very problem under a different name. The socialist calculation debate asked whether experts could steer an economy from the center, and the answer given by Ludwig von Mises in a famous 1920 essay, and elaborated on by Hayek, was that they could not. The planners’ problem was information, not intelligence. The knowledge they needed was scattered across millions of individuals and could never be assembled in one place, while prices assembled it automatically.

A harness is a small planned economy. One designer decides in advance which actions are permitted, for agents whose circumstances the designer cannot fully foresee. That works tolerably when the swarm is small and the tasks are known. As agents multiply and their tasks grow less predictable, the harness designer’s position starts to resemble the central planner’s, and the calculation debate tells us how that will end.

Since agents currently have none of this infrastructure, harnesses may be the only tool that work right now, and they have a real advantage in the short term. Eventually, though, as agents begin dealing with one another on a large scale, norms and practices may start to emerge that facilitate good behavior. The practical task at that point is to watch what develops and to codify the best practices that prove themselves. That is how commercial law came about. Merchants worked out their own customs long before any legislature wrote them down.

Let Order Emerge

In a recent article, Tyler Cowen and Sonia Farrell Pearson propose requiring AI agents to hold capital. Their concern is the “untethered” agent, one whose actions cannot be traced back to any legally accountable human. Strictly speaking, a capital requirement is another harness—a rule somebody designs and imposes from outside, verifiable as a step in a process. It is a worthwhile experiment, and sometimes the legal regime really does have to come first. Corporate charters, limited liability, and property registries all enabled the markets that grew on top of them. Those markets could not have functioned without the legitimacy the law supplied.

But the problem here is squishier. Before an agent can be required to hold capital, it needs an identity that lasts long enough to own something, and the law has to recognize the ownership. Nobody sat down and designed those features of human commerce. They built up over centuries of people actually trading with one another. Agent markets will probably develop in the same way. Agents will start transacting, workable practices will appear, and the law will absorb the ones that stick.

A recent Lawfare article by Nathan Darmon and Tom Reed sketches what that absorption mechanism might look like. AI companies have already written constitutions for their models, but as Darmon and Reed observe, “there is no institutionalized mechanism to gradually refine rules over time, absent a major public backlash.” Their proposal is to give the constitutions courts. These might consist of small internal tribunals that hear hard cases and publish reasoned opinions, which then build a body of precedent that feeds back into how models are trained. If agent markets are going to generate their own customs, something like Darmon and Reed’s courts could be where those customs get formalized into doctrine, the way the common law absorbs the customs of market participants.

For practical purposes, none of this is a criticism of building harnesses. Krier is right that alignment is a process rather than an end state. In the coming years, don’t be surprised if the norms and customs of agent markets develop faster than the verifier stacks. Constitutions are a great achievement of institutional design. But most of the good behavior in a commercial society is enforced by neither constitutions nor character. It is enforced by customers who can walk away. This mechanism has a longer track record than any harness. And it may be the best hope we have for aligned AI.

Tyler Cowen and Sonia Farrell Pearson published an essay on August 18 asking whether AI agents that nobody controls could be governed by making them hold capital. Untethered is their term for an agent nobody controls:

By “untethered” – a central concept in this essay – we mean that there is no meaningful or actionable way to trace the actions back to a legally accountable human or institutional entity.

This is less hypothetical than it sounds. Cowen and Pearson point out that thousands of hobbyists already run open-source bots on their own machines, and an agent can end up untethered by accident too if its creator dies or disappears, or its nominal owner is a shell company that offers no real defendant.

Their essay is timely, given Javier Milei has proposed letting AI agents run their own companies under Argentine law. Yet critics like Yuval Noah Harari object that the sanctions that discipline human executives, like jail, mean nothing to an algorithm. Cowen and Pearson suggest that if agents held capital, bad behavior could cost them something, similar to the way capital requirements make banks bear their own losses and behave more cautiously.

The essay has a tension at its center, however, and to the authors’ credit they more or less concede it. How do you require anything of an entity whose defining feature is that there is no one to require anything of? Their answer is that the requirement never applies to the agent directly. It applies to everyone the agent needs to deal with.

Enforcement through counterparties

Even an untethered agent has to buy compute, move money, and enter contracts with someone. So the rule is imposed on those counterparties, who in practice will be entities like compute providers, payment companies, and other firms that interact with agents. They will face liability if they deal with an unregistered, uncapitalized one. The essay uses a maritime analogy, borrowed from Shruti Rajagopalan:

An agent that “cannot present a registration” is treated as a “stateless vessel, presumptively unlawful, and every compliant provider may refuse it”.

Nobody compels a stateless ship to flag itself. They just make it very hard to dock anywhere. Seen this way, capitalization is the admission fee to the legitimate economy. An agent remains free to stay unregistered and uncapitalized; it just gets progressively walled out of the transactions that make it useful. Because an untethered agent cannot be forced into a legal arrangement, “the goal becomes making it as difficult as possible to operate without one.”

In defense of chokepoints

Enforcement through financial intermediaries has a bad reputation among some. Operation Choke Point, launched by the Justice Department in 2013, put pressure on banks to drop payday lenders, gun dealers, and other lawful businesses the Obama administration found distasteful. The DOJ shut down the effort in 2017 after acknowledging it was hurting legitimate businesses. The sequel, which the crypto industry dubbed Operation Choke Point 2.0, involved FDIC “pause” letters urging banks to hold off on serving crypto clients.

The defect in those episodes was the target, however, not the technique. Regulators went after businesses that were legal but politically disfavored, and they did it through bank supervision precisely because no statute would have authorized going after them directly. That is an end run, and it deserved the backlash it got.

But reaching bad actors through the intermediaries they depend on is not, in itself, illegitimate. It is a standard tool of enforcement, it can be strategic, and sometimes it is the only tool available. The untethered agent is a case in point. If there is no owner to fine, no body to jail, and no headquarters to raid, the intermediary may be the only door the law has left to knock on. What separates a defensible chokepoint regime from Choke Point 1.0 and 2.0 is that the rule should target conduct no one defends and do so openly, through law, rather than through a quiet word from an examiner.

A financial transaction tax on AI agents could rest on similar logic. Rather than use it as a weapon to target an unpopular industry, such a tax could be a way to discourage specific unwanted behaviors. Email spam exists because sending a message costs nothing, so sending ten million costs nothing too. A penny per message would end the practice without anyone having to distinguish spam from ordinary email. The reason no one has ever taxed email is that it has no chokepoint to collect at. It runs on an open protocol and much of the traffic comes from hijacked machines whose owners would be the ones billed. What ended up policing spam instead was the emergence of private chokepoints. A handful of providers now handle most inboxes, and they have learned how to filter and block delivery.

AI agents present the same volume problem but without email’s collection problem, because agent transactions already pass through billable gateways. A small per-transaction fee, collected from compute providers and payment processors, would be trivial at normal volumes but prohibitive at the machine-scale volumes abuse requires.

The evasion problem

Cowen and Pearson admit there is an enforcement gap, however. Untethered agents can still rely on private hardware, stolen credentials, and crypto holdings to pursue their end goals outside the formal legal system.

All leave room to operate outside of our infrastructure, so agents with a reason to evade official channels will likely be able to do so. What we’re catching, then, is the agent that was mostly law-abiding to begin with.

This is a problem for their proposal, but it is also a standard result for any regulatory regime. Compliance costs fall only on the compliant. Those who operate in the shadows always escape them.

This is a problem for their proposal, but it is a problem regulators deal with already, especially in finance. Know-Your-Customer rules are supposed to give every account a documented owner, but an identity can still be faked, in which case there is no account owner to hold accountable. Anonymous shell companies work similarly, with no assets and no human being behind them. Legal entities that exist on paper but are empty in reality are as old as the corporate form itself, but nobody treats these practices as a reason to abandon the rules. The compliant bear the costs, some bad actors slip through, and a regime is judged on whether it improves behavior at the margin. There is nothing AI-specific about this challenge, although AI could make the problem much bigger.

The regime Cowen and Pearson endorse seems to presume untethered agents want something the legitimate economy supplies. It also assumes that capital accumulation is a goal agents pursue, in part because it is instrumental to achieving other goals. These seem like reasonable assumptions, though neither will universally be true.

Untethered agents, limited policy options

The takeaway, then, is that capitalization is a solution for agents with enough dependence on lawful infrastructure that an indirect requirement reaches them. That may well describe most agents most of the time, just as fines and penalties incentivize most people through the accounts and property they depend on. But if fully untethered agents become common—and we genuinely don’t know if that will be the case—then by the essay’s own definition, they will be mostly outside the reach of policy solutions like this one.

Late last month, The Economist magazine published an essay by Gita Gopinath, Pierre-Olivier Gourinchas and Hélène Rey in which the authors argued that demands for China to revalue the yuan get “both the diagnosis and the cure wrong.” Michael Pettis, a finance professor who has spent two decades in Beijing arguing that countries' trade balances are set by how much they save and invest, published a response on his Substack the next day. By August 1, Gopinath was on X writing that after reading his reply she “wondered if Michael had read our piece or some other piece.”

This has the tone of a real fight. Read the two essays side by side, however, and what stands out is actually how much the combatants agree on. They agree on the problem, the solution, and even the politics of getting there. The disagreements concern which policy lever to pull first and how a given lever transmits to the trade balance. Those questions matter, but they are narrower than the public exchange suggests. With global imbalances widening again, the shared ground is the useful place to focus.

The debate

The Economist essay is a call against what its authors term “a comfortable ritual in international economic diplomacy.” Whenever imbalances flare up, the conversation gravitates to the exchange rate. Their central claim is that the yuan’s weakness is an effect of Chinese domestic policy rather than an independent cause of the country’s trade surplus:

A country that suppresses household consumption while simultaneously facing a collapse of property investment will run persistent current-account surpluses and will, other things equal, have a weak currency. … The currency is misaligned because the underlying policy mix produces too much or too little saving. The exchange rate is a symptom, not the disease.

From there they run a thought experiment. Suppose China revalued tomorrow, sharply and by fiat, with no change in its underlying policies. Little would happen to exports, since most Chinese export prices are set in dollars and adjust only slowly. Meanwhile, “the stronger yuan would further exacerbate China’s deflation problem and reduce its overall demand,” which is the opposite of what rebalancing requires. A nominal appreciation unsupported by macroeconomic change, they argue, would quickly erode into a real depreciation through deflation.

Pettis grants the causal point but contests the policy conclusion:

Where they are mistaken is in failing to recognize that currency appreciation remains one of the most direct ways of reversing the domestic income imbalance that generates that surplus.

According to Pettis, “China’s trade surplus is better understood as the external manifestation of an internal imbalance in income distribution than the other way around.” On Pettis’s account, Chinese households retain an unusually low share of national income, so they consume an unusually low share of what China produces, and the difference is exported.

That premise is not in dispute. The Economist authors’ own prescription is to raise the Chinese household income share. A stronger currency works on that margin directly, because “a stronger renminbi raises the purchasing power of Chinese households by making imported goods and services cheaper in domestic currency terms,” Pettis writes. Appreciation, on this view, is one member of a family of policies (e.g., larger fiscal transfers, a stronger safety net) that shift income toward households and thereby shrink the surplus.

A tally of agreement

Gopinath was unpersuaded, as her August 1 thread on X made clear. But set the rhetoric aside and look at what the two sides jointly agree on. Both believe the yuan is too weak. The Economist authors say it plainly (“To be clear, the Chinese yuan is no doubt undervalued”). Both locate the root cause of China’s surplus in domestic policies that suppress the household share of income and consumption. Here is the Economist authors’ own prescription:

What China needs to do—and what is in its own long-term interest, given its ageing population and an investment model overly reliant on the tradable sector—is to raise the share of household income in GDP, expand social insurance so that families feel able to spend, and stop financing tradable-sector expansion at the expense of consumption.

That paragraph could have appeared under Pettis’s byline. It is close to a summary of the argument he has been making since at least his 2013 book, The Great Rebalancing. Both sides also agree that the conventional rationale for currency revaluation—that a pricier yuan makes Chinese exports more expensive and imports cheaper, so trade volumes adjust—is not the most important mechanism. The Economist essay dismisses what it calls “exchange-rate gimmicks,” while for Pettis appreciation matters only insofar as it shifts income toward households. He goes further, “in fact any domestic policy that causes a shift in the distribution of income domestically is also, in effect, a trade policy.”

The two camps even expect the same endgame. After discussing ways to raise the household share of GDP, the Economist authors note, “Do those things, and a real appreciation of the yuan will follow.” Pettis reverses the order of operations, as his title makes clear: currency revaluation is income rebalancing. On his view appreciation does not set rebalancing in motion at some later stage; it is itself one of the transfers to households that rebalancing consists of. One side treats the currency as the dependent variable; the other treats it as an instrument.

The endpoint, a higher household income share alongside a stronger real yuan and a smaller surplus, is common to both. And each acknowledges that rebalancing is politically hard for the same reason. The Economist essay warns that “[w]hen you push for an exchange-rate adjustment, you ask for conflict,” while Pettis observes that “[e]very successful rebalancing policy necessarily transfers income away from businesses or governments and toward households.” Someone powerful pays along the way.

What remains in dispute is the transmission mechanism. Gopinath and her coauthors analyze appreciation through the trade-price channel. A stronger yuan is supposed to make Chinese goods dearer abroad and foreign goods cheaper at home, so quantities adjust. They find that channel weak, because most Chinese exports are invoiced in dollars and those prices adjust only slowly. This strategy is also risky, according to the authors, because falling yuan prices would deepen China’s deflation and suppress the domestic demand changes rebalancing requires. Pettis analyzes the same appreciation through the income channel. Cheaper imports raise the real purchasing power of yuan-earning households while trimming exporters’ revenues, transferring income from businesses to the households whose weak consumption is the source of the surplus.

The two camps are reading the same set of factors as a deflationary drag in one case and as a real income gain in the other. Which effect dominates is an empirical question about magnitudes. That would be a more productive thing to argue about.

What about America?

The Economist authors write that “America will also need to tackle its unsustainable fiscal policy, as well as its persistently low private savings.” Pettis is skeptical that fiscal consolidation would change the US external position, because he sees the capital inflows as supply-driven, pushed by excess savings abroad rather than pulled by Treasury issuance. If Washington borrowed less, he argues,

Foreign investors would simply buy different US assets. The composition of American liabilities might change, but the underlying capital inflow – and therefore the corresponding current account deficit – would remain largely intact.

He may well be right about much of the flow. A share of the foreign money now parked in Treasuries would indeed migrate into corporate bonds, equities and other dollar assets, leaving the trade deficit somewhat unchanged. But “largely intact” assumes those assets are close substitutes for Treasuries, and for many of the marginal holders they are not. Treasuries are the deepest and most liquid safe asset in the world. Corporate and municipal debt meanwhile carry credit risk and have thinner markets. For a reserve manager or a foreign institution seeking safety, a German bund may be a closer substitute for a Treasury than a US corporate bond. Shrink the supply of Treasuries and at least some of that money would leave the dollar bloc altogether.

The current account equals national saving minus domestic investment, and a smaller fiscal deficit raises public saving directly. The identity alone settles nothing—private saving and investment would respond, and the response would offset part of the improvement—but a partial offset is not the full offset Pettis’s argument requires. Consolidation would not eliminate the US trade deficit, but it would likely shrink it, and unlike demands for changes in Beijing currency policy, it is entirely within Washington’s control. Fiscal consolidation is also worth doing for reasons that have nothing to do with trade.

Why this debate is back

Global current account balances—the sum of the world’s absolute surpluses and deficits—peaked near 6% of world GDP before the financial crisis, narrowed to about 3% by 2018, and stayed near that level for years. In 2024, they jumped to 3.6% of world GDP. The IMF’s staff assessment attributes about two-thirds of that widening to excess balances, meaning gaps beyond what economic fundamentals would justify, and calls the recent increase in excess balances the largest in a decade.

Meanwhile, China’s goods trade surplus reached a record $992 billion in 2024 and then, despite Trump’s tariffs, roughly $1.2 trillion in 2025, according to Chinese customs data. The IMF’s July 2026 External Sector Report finds that global balances “widened further in 2025,” continuing the post-pandemic trend.

I made the case in a Forbes piece in May that these imbalances deserve more attention than they have received since the financial crisis-era faded. What the Gopinath–Pettis exchange shows is that there is now a fair amount of consensus that there is a problem, and on where the problem comes from, too. Arguably the largest single cause is a set of Chinese policies that hold down the household share of income, with US fiscal deficits contributing from the other side. The disagreements are over how to get from here to better outcomes, and over the mechanisms by which good policy produces them.

If Beijing expands social insurance and transfers while Washington’s fiscal path tightens—or if neither happens—the IMF’s external sector assessments will record the result, showing whether the excess component of global balances narrows. Also worth watching is whether the yuan appreciates on its own, as both camps predict it eventually must. A debate in which everyone agrees on the disease and on what a cure would accomplish, but argues over the prescription, is close to a consensus by the standards of international economics.

Alex Tabarrok published a characteristically sharp post at Marginal Revolution on July 28, skewering Britain’s attempt to have employment tribunals decide whether a shelf-stacker’s job is of “equal value” to a warehouse worker’s. His thought experiment imagines Parliament convening a tribunal to rule on whether apples and oranges are truly of equal worth, and then walks through everything the tribunal would need to know. This includes the opportunity cost of Florida grove land, the interest rates embedded in orchard investments, and the fact that apples keep while oranges rot, among other things. His conclusion is that pricing apples against oranges requires an entire general-equilibrium system, which is exactly what the price system computes and no tribunal can.

On the core claim, Tabarrok is right, and the UK evidence he marshals is damning. His leading example involves Tesco, Britain’s biggest grocer. Since 2018, tens of thousands of mostly female shop-floor workers have claimed under the UK Equality Act that their jobs are of equal value to those of the better-paid, mostly male staff in Tesco’s distribution centers. The law, they argue, therefore entitles them to the same pay. Eight years on, the litigation has generated more than 900 pages of judgments, the most recent from the Court of Appeal in May 2026, but the independent experts have still not begun the report that would actually compare the jobs.

The Category Error Argument

Tabarrok’s post reaches past the UK example to make a broader philosophical claim. He argues that demanding prices (or in the Tesco case wages) be just is a category error, “like suing the weather.” Tabarrok’s point comes from Hayek’s The Mirage of Social Justice, and he states it well:

Hayek nailed it in The Mirage of Social Justice: justice is about conduct — how one person treats another. An employer who defrauds his workers, an employee who steals from the till, a product sold under false pretenses — condemn them, take them to court. But the pattern of prices that emerges from millions of voluntary trades is nobody’s conduct. No one chose it, no one designed it, no one can be guilty of it. The constellation of prices is, in Ferguson’s phrase, the result of human action but not of human design. Demanding that prices be just is a category error, like suing the weather. Prices don’t grade our merit; they guide our actions. Ask them to do the first and they can no longer do the second.

In other words, prices emerge from a decentralized process that aggregates dispersed knowledge, and that process is indifferent to human conceptions of equity. Nobody sets the relative wage of nurses and plumbers, so nobody can be indicted for it. As Tabarrok puts it in the same post:

A wage is not a grade on your character or a measure of your worth as a human being. A wage is a price — a report on how scarce your skills are relative to the desires of people you will never meet.

There is real force to this account, and it helps explain why the UK’s Equality Act tribunals are probably doomed. Scoring jobs by their intrinsic effort, skill, and responsibility is, as Tabarrok notes, “the labor theory of value applied to labor.” That approach asks what a job is rather than what it is scarce relative to, or how desired it is, which may be why eight years of adjudication have produced no answer.

No one's to blame for the weather. But it can still be a bad day.

But notice what the category error argument actually requires. It works through Hayek’s particular conception of justice, where justice is defined as a property of conduct, of how one person treats another. Under that conception, the pattern of prices cannot be unjust because it is nobody’s doing. Even that assertion can be pushed back on, however. Prices emerge within institutions that people do design and revise, including the law, property rights, taxes, and so on.

But even accepting Hayek’s claim that prices are no one’s handiwork, conceptions of justice are personal, and mine may differ from yours, and both of our definitions may differ from Hayek’s or Tabarrok’s. If a person’s standard of justice is Rawlsian, for example—an arrangement is just when it works to the advantage of the least well-off—then when that person calls prices unjust, no category error has been committed. The complaint is a claim about how outcomes measure up to a particular standard.

Hayek, for what it is worth, did not treat Rawls’s theory as a category error. In fact, in the preface to the very volume Tabarrok cites, Hayek explained that he would not engage Rawls’s A Theory of Justice at length “because the differences between us seemed more verbal than substantial.”

And a Rawlsian conception of justice is just one standard a person could apply. Any normative goal can serve as a benchmark. Pick the goal, and whether current prices and allocations are consistent with it becomes a factual question. Statements of that kind are objective and scientific. They assess a relationship between an outcome and a standard, and they can be right or wrong, even though the underlying goal is normative.

Economists run this procedure every day with efficiency. An allocation is efficient when no rearrangement of resources would increase their total value to society, and a price is efficient when it equals the resource’s shadow value—what the marginal unit would be worth in the present moment if every relevant scarcity and spillover were accounted for. Far from being a philosopher’s imposition, efficiency describes what markets do when they work, which makes it the natural yardstick for noticing when they don’t work well and prices are not reflecting their true social value.

Signals Can Be Wrong

Tabarrok has elsewhere argued that a price is a signal wrapped up in an incentive. A signal, however, can be wrong. When a factory’s price for its output omits the cost of the pollution it emits, the signal reaching every downstream user is distorted, and the incentive wrapped around it points in the wrong direction. This is what economists mean by market failure. Prices deviate from their shadow values, guiding resources toward lower-valued uses than would be optimal. Judging a price against its shadow value is no more a category error than asking whether the weather matched what the almanac promised. Prices are guides, as Tabarrok says. But sometimes guides lead us astray.

One reply available to Tabarrok is that this argument seems to bless the very tribunals whose failure opened this post. However, identifying a price that is inconsistent with a benchmark is a far easier problem than computing and decreeing the correct one from scratch. A doctor can diagnose a fever without specifying the patient's ideal temperature at every hour of the day. This is why the remedy for a distorted price is usually to remove the distortion rather than to legislate a new number.

Its diagnosis rests on a benchmark—the intrinsic value of a job—that assumes value is an inherent property of the work rather than a function of scarcity, and its remedy—pay parity by decree—sets a price rather than removing whatever distortion might explain the gap in wages.

The Marvel Can Still Be Judged

None of this diminishes what Hayek referred to as the marvel of the price system. In the same 1945 essay Tabarrok cites, Hayek wrote:

I have deliberately used the word “marvel” to shock the reader out of the complacency with which we often take the working of this mechanism for granted. I am convinced that if it were the result of deliberate human design, and if the people guided by the price changes understood that their decisions have significance far beyond their immediate aim, this mechanism would have been acclaimed as one of the greatest triumphs of the human mind.

The price system is indeed a marvel, and Tabarrok is right that it is the only institution yet established that can credibly compare the value of apples and oranges. But a marvel can still be judged, and judging it against the one standard native to it—how well it allocates resources—is a core task of economics rather than a confusion about categories. True, the weather is not plotting against us. But it can be graded as consistent or inconsistent with any number of benchmarks, so long as they have empirical implications.

In that sense Hayek was wrong, or at least he proved less than he claimed. That nobody designed the constellation of prices means only that nobody can be blamed for it. It does not place the pattern beyond evaluation. A marvel nobody made can still be judged against whatever normative standard a person brings to it, including justice.

As of July 21, Substack readers can now run any newly published post, note, or comment through a built-in AI detector, powered by a detection company called Pangram. After clicking a button, users receive a percentage estimate of how much of the text was human-written and how much was AI-assisted. Readers weary of machine-generated slop have cheered the feature, while some writers who use AI somewhere in their process have called it a “witch hunt.” Both reactions treat “did a machine touch this prose?” as the question that matters. A better question is whether a real author with a real argument stands behind the piece, because most readers were never paying for the typing.

Writing is a production process

An essay is the output of several distinct stages: forming the idea, doing the research, constructing the argument, drafting the prose, and editing the result. These stages have different costs for different writers. For some, the drafting comes naturally and the thinking is expensive. For many others—myself included—the first draft is the hardest part. It’s a slog of turning notes and a mental outline into functional paragraphs, followed by round after round of editing.

AI has sharply lowered the cost of that stage. The economically sensible thing to do when there is a change in relative prices is to reallocate your time toward the stages where you still hold a comparative advantage. For a policy writer, that means the ideas become relatively more valuable, since competent expository prose is becoming a commodity. A writer who insists on hand-drafting every sentence of a research report or blog post is making the same choice as a tailor who insists on hand-stitching every seam. This is defensible as a craft preference, but not always a mark of quality the customer can detect in the finished garment. And there is a real cost in terms of foregone output. The extra time taken to hand stitch means other garments go unmade.

Mack Collier, who writes a Substack called Backstage Pass, put this plainly in a note quoted by 404 Media:

I’m not going to apologize for using AI in the creation process. I wrote for 20 years without AI, I could do it again if I wanted to. My output would fall, my posts would be less structured, and it would be more obvious that they needed a good editor. Using AI improves the overall quality of my writing. That’s why I use it.

There is also nothing new about separating the byline from the drafting. Judicial opinions are drafted by clerks. Presidential speeches are drafted by speechwriters. CEO shareholder letters pass through communications staff. In each case, the named author supplies the ideas and the accountability, and readers generally don’t demand a forensic accounting of who typed every sentence.

What the detector actually measures

Pangram produces a percentage estimate of how much of a text was human-written and how much was AI-assisted, based on statistical patterns. The giveaways tend to be text where every next word is the statistically likely one, where sentence after sentence has the same length and shape, and where there is a fondness for words like “delve,” “tapestry,” and “underscore” that people rarely reach for on their own.

Pangram’s CEO, Max Spero, told 404 Media the company estimates its false positive rate at roughly one in 10,000. Even taking that at face value, one error per 10,000 means that for every few million human-written posts readers scan, hundreds will get publicly scored as machine-written. Alice Lemee, a ghostwriter and writing coach, warned that “all it takes is one false accusation for a writer to have their reputation almost irreversibly tarnished.”

Substack writers do have some recourse, but it is limited to disabling detection on their posts, attaching a statement explaining their process, or clicking a “Report detection error” button that asks Substack to review the AI detector’s verdict.

In a July report drawing on roughly a million posts scanned through its browser extension, Pangram found that more than 40 percent of longform LinkedIn posts flagged as fully AI-generated, that nearly half of X articles were either fully AI-generated or AI-assisted, and that Substack—at just over a fifth of posts—had the lowest combined rate of the longform platforms it measured. Thus, the platform with the least AI writing is the first to deploy a detector against it.

That makes economic sense once you consider what each platform sells. LinkedIn and X sell access to an endless feed of news and commentary, where perhaps nobody expects much humanity behind the posts. Substack sells ongoing subscriptions to particular named writers, so its business depends on readers believing there is a specific person behind the byline. Introducing the scanner seems to fit with Substack’s business model.

Sam Illingworth, captured the odd irony behind the whole exercise:

We have now built the mirror image. Pangram is a real machine, trained on vast amounts of text to make a probabilistic guess, and we have pointed it at your writing to work out whether a person is hidden inside… To decide if there is a human on the other end, Substack asks a machine.

In other words, Substack’s remedy for distrust of machine judgment is another machine’s judgment. There is a Goodhart’s law problem lurking here as well. Once a detector score becomes the thing readers check, writers will optimize their prose to avoid detection rather than for the reader. The statistical fingerprints the tool relies on are exactly the features a motivated writer can learn to avoid. Detection scores will then certify nothing except that the author cared enough to launder the text.

The voice exception

One honest caveat is that some writers really are selling the sentences more than the ideas. When it comes to George Will’s ornate sentences or Paul Krugman’s snarky and condescending tone, readers really do subscribe to the voice as much as to the analysis. Running that kind of writing through an AI model would flatten it toward the statistical mean, which is precisely what the model is trained to produce. If your voice is genuinely inimitable, AI assistance is a tax on your differentiation, and you should use it sparingly if at all.

But writers in that category are rare, perhaps a handful of columnists per generation. Most academic and policy writing sits at the other end of the spectrum. Journal articles and think tank reports are written in a detached, neutral tone that deliberately suppresses voice in favor of precision and objectivity. The reader of a cost-benefit analysis wants accuracy. The sentence rhythm is merely incidental. For that kind of work, which is to say for most professional nonfiction, machine-assisted text that has been checked and owned by its author is simply fine.

Even the writers most invested in hand-crafted prose are not abstaining completely. Tyler Cowen—who predicted in The Free Press that within a couple of years “people will not care if a chatbot wrote what they’re reading” — disclosed in the same column that while he does not use AI to draft, he does use it to fact-check and proofread, and that he expects to end up “as the dinosaur in the room” as other writers produce better-informed work with more machine help. But fact-checking and proofreading are stages of the production process like any other, so it is hard to see why drafting should be the one stage a machine must never touch. At this point, the writer who does not use AI at any stage in the writing process barely exists among serious professionals.

Adapt or Fall Behind

For practical purposes, my advice runs opposite to the mood of the moment. Notice that the loudest opposition to AI-assisted writing typically comes from the most established writers. However, this is exactly what you would expect. They built their reputations and audiences in an era when producing high-quality prose unaided was the only game in town. Incumbents rarely welcome a technology that devalues the skill that made them incumbents. Their position is untenable in the long run, however. If you write about policy or scholarship for a living and you are not using AI somewhere in your process, you are paying more for your inputs than your competitors, and you will eventually fall behind.

I suspect Substack’s scan counts will decline in popularity within a couple of years as readers stop bothering to check. Meanwhile, newsletters that openly disclose AI assistance will hold on to their subscribers. A machine can now tell you whether a machine touched the prose. It still cannot tell you whether anyone had anything to say.

Noah Smith published a thoughtful essay over the weekend asking what more intelligence will do for the economy. His answer is that even if intelligence hits diminishing returns, AI can still deliver large productivity gains through channels that have little to do with being smarter than any individual human. The most interesting of these channels is what he calls distributed “tacit knowledge.”

Let’s start with his example. The German firm Zeiss makes what Smith calls the best glass on the planet, including the mirrors inside ASML’s chipmaking machines. How? Smith’s answer:

Zeiss’ technology basically can’t be stolen, because it’s tacit and distributed. It consists of a vast number of little tricks and techniques that a huge number of individual employees use on a daily basis. These people don’t always even realize all those little things they’re doing that make the glass come out so good. And each employee knows a different set of tricks and techniques. The knowledge exists at the level of the organization itself, and is thus very hard to steal or recreate.

Tacit knowledge is the knowledge embodied in skill and practice that cannot be fully articulated even by the person who has it. It is a concept usually credited to the chemist-turned-philosopher Michael Polanyi, whose formulation was that “we can know more than we can tell.” It is also closely tied to the dispersed-knowledge argument at the heart of F.A. Hayek’s case against central planning. But then Smith goes further, and this is where things get interesting:

Suppose American rare earth refiners give their employees a bunch of equipment to record everything they do — smart glasses, gloves, and so on — in addition to sensors distributed throughout their plants. AI will be able to synthesize all that information and very rapidly suggest small ways to improve the production process... In other words, in the age of AI, distributed tacit knowledge might not be nearly as big of a barrier to technological diffusion.

So Smith thinks if we strap enough sensors on enough workers, then the tacit knowledge becomes explicit. Knowledge that formerly lived only in the fingertips of a Zeiss glass polisher becomes data that is recordable and transferable. Smith presents this as a growth scenario, and a plausible one. What he does not mention is that this is also the premise on which successive attempts to rehabilitate central planning have rested.

The argument Smith has made elsewhere

Smith knows the Hayekian argument well and has presented it elsewhere approvingly. In a 2024 essay on liberal democracy, for example, he summarized it this way: “people themselves have a lot more information about both what they want, and about how to use productive resources efficiently, than a central planner does,” and the function of markets is to aggregate “those dispersed small bits of useful knowledge in order to better inform production decisions.”

He even quoted this famous passage from Hayek’s “The Use of Knowledge in Society”:

The peculiar character of the problem of a rational economic order is determined precisely by the fact that the knowledge of the circumstances of which we must make use never exists in concentrated or integrated form but solely as the dispersed bits of incomplete and frequently contradictory knowledge which all the separate individuals possess... it is a problem of the utilization of knowledge which is not given to anyone in its totality.

Hold the two claims side by side. In the 2024 essay, dispersed knowledge cannot be concentrated, which is why command economies fail and market economies succeed. In the 2026 essay, dispersed tacit knowledge can be concentrated—recorded by smart glasses and plant sensors and synthesized by a model—which is why AI could accelerate growth. This is not a criticism of Smith. I am merely pointing out that if AI can extract what Zeiss’s workforce knows without any of those workers being able to articulate it, then the “knowledge problem” that Hayek and others treat as a decisive argument against planning starts to look like an engineering problem awaiting better instrumentation. A central planner with the same sensor network would seem to be back in business.

What the calculation debate actually established

This is precisely the move that Peter Boettke and his coauthors have spent years warning against. In a recent short monograph on the century-long argument Ludwig von Mises started in 1920, Boettke (along with coauthors Rosolino Candela and Tegan Truitt), traces how each generation of socialists has reframed the calculation problem as a computation problem. This first happened with Walrasian equations, then with linear programming, then with mainframe computers, and now with machine learning and AI. In a companion paper on “technosocialism”, Boettke and Candela state the issue this way:

[T]echnosocialism represents a restatement of the case for market socialism, which incorrectly framed the “solution” to economic calculation under socialism as one of computing data, rather than the discovery of context-specific knowledge that only emerges through the exchange of property rights.

And more pointedly:

[T]echnosocialism cannot replace the competitive discovery process that takes place only within the context of the market, one in which residual claimants are the locus of decision-making, responsible not only for their correct decision-making communicated in the form of monetary profits, but more importantly their errors communicated as losses.

The point is that much of the relevant knowledge does not exist prior to the market process that elicits it. What a resource is worth in its next-best use is not a fact sitting in someone’s head waiting to be recorded; it is generated by acts of choice, bidding, and exchange, and it dissipates as conditions change. No sensor records a valuation that has not yet been formed.

There is a way to reconcile Smith’s two positions. One can hold that production knowledge—the Polanyi-style skill embodied in how a worker polishes glass or titrates a solvent—is in principle observable and therefore partially capturable by instrumentation, while market knowledge—the contextual, ephemeral knowledge of preferences that Hayek emphasized—is not, because it does not exist externally until people make choices, so there is nothing for a sensor to record. On that reading, AI can capture subtle details about technique without touching the calculation problem.

This distinction may well be right. But it concedes something important. Even within the factory, the sensors are capturing behavior, not the kind of tacit knowledge that underpins markets. The model learns what the polisher’s hands did, now what the polisher knows about the client this mirror is for. How much the gap between this kind of tacit knowledge and behavior matters is an empirical question, and nobody knows the answer yet.

Does AI have tacit knowledge of its own?

If tacit knowledge is knowledge a system uses but cannot articulate, then large AI models are arguably the most concentrated repositories of tacit knowledge ever built. A frontier model’s competence is distributed across billions of weights. No one, including its creators, can fully state the rules by which it writes a proof or a paragraph. Smith himself describes the concept well. His notion of “cloud laws,” which are regularities “too diffuse and complex for an individual human being to either intuit or communicate,” but which AI can nonetheless exploit, can be seen as a description of tacit knowledge residing in a machine.

This slightly reframes the interpretability project in AI research. Interpretability is the attempt to make machine knowledge explicit—to open the black box and articulate what the weights encode. It is the same extraction problem Smith wants to run on the rare-earth refinery floor. And there is little reason to expect it to end in full transparency. Just as we may never fully emulate human capabilities because some human knowledge is tacit all the way down, we may never fully interpret machine capabilities for the same reason. The economy may end up as an exchange between two kinds of opaque knowers—humans who cannot say what they know, and machines that cannot either.

For practical purposes, this may not matter much, and this is where Smith’s growth scenario survives in modified form. Automating a job does not require a complete theory of the worker any more than using a model requires a complete theory of its weights. Recording people’s work habits may replicate enough of human ability to be transformative, in the same way that partial interpretability may be enough to deploy models responsibly. “Good-enough” approximation is a perfectly respectable engineering standard. What it is not is a refutation of Hayek. The residual that the sensors miss is small enough to ignore when you are optimizing a production line, but it may be the thing that matters most when you are trying to plan an economy.

The test over the next decade will be whether firms that instrument their employees can turn those recordings into anything close to Zeiss-grade capability, and whether that helps laggard firms close the productivity gap with frontier firms. If that happens, Smith will have been right about the benefits of AI for economic growth. But the calculation debate will be exactly where Mises left it in 1920.

Last week, the White House Office of Science and Technology Policy released a report to the President titled, Science: A New Golden Age. Among its recommendations is a call to “enforce reproducibility, transparency, data sharing, and falsifiability across all federally funded research.” The report also proposes rewarding researchers “who replicate or disprove influential scientific results.”

Ideas that spent decades as academic theories about how science ought to work are starting to become the operating assumptions of the federal government. Falsifiability—the principle that a genuinely scientific claim must specify what evidence would prove it wrong—has moved from the philosophy seminar into compliance language for research grants. And if the report’s recommendations are adopted, researchers may soon be paid for proving theories wrong, an outcome the current system hardly rewards at all.

The conventional account treats this evolution of science as a recent story. It begins in 2005, when the epidemiologist John Ioannidis published “Why Most Published Research Findings Are False”, arguing that small samples and the race for striking results make false positives the norm in many fields. In 2015, the Open Science Collaboration re-ran a hundred published psychology experiments, and most of the results did not hold up. Just this spring, a years-long project published in Nature retested social-science findings and could confirm only about half of them.

Out of this “replication crisis” emerged the field of metascience—the scientific study of science itself—applying the tools of research to how research gets funded, conducted, and published. Metascientists proposed a set of reforms that are becoming standard vocabulary in research policy. Preregistration requires researchers to publicly specify their hypotheses and methods before collecting data, so results cannot be quietly reshaped after the fact. Open data policies require that the underlying numbers be made available for others to check. A dedicated UK metascience unit is now up and running in this spirit, along with funding experiments at NSF and NIH.

But the intellectual history runs deeper than the past two decades. Between the 1940s and the 1960s, a debate played out among a handful of scholars over what science is and what it should be. At the time, the debate was almost entirely theoretical. But today, governments are applying the lessons from that period at a scale larger than ever before.

What Science Should Be: Merton and Popper

Start with Robert K. Merton, the sociologist who argued in 1942 that science runs on norms, not just method. Merton identified four rules—he called them “institutional imperatives”—that hold the scientific enterprise together. These are universalism (claims are judged on their merits, not on the status of the person making them), communism (findings belong to the community, not the discoverer), disinterestedness (the reward structure favors advancing knowledge over advancing oneself), and organized skepticism (every claim is open to scrutiny, all of the time).

Merton's deeper point is that science is an institution, and like any institution, it holds together only as long as its members' incentives align with its core norms. The modern metascience reform agenda is highly Mertonian in this respect. Open data mandates operationalize his communism, critiques of publish-or-perish incentives seek to rehabilitate disinterestedness, and replication audits institutionalize organized skepticism.

Karl Popper supplied the sharpest single standard for demarcating science from everything else. “The criterion of the scientific status of a theory is its falsifiability, or refutability, or testability,” he wrote in Conjectures and Refutations. On Popper’s account, a scientist’s job is to state in advance what evidence would refute the theory, and then to go looking for it. The reforms metascientists favor most are Popperian machinery. Preregistration commits a researcher to a hypothesis before the data can tempt them toward a different one. Registered reports, where journals also pre-commit to an acceptance, go further. Journals review and accept papers on the strength of the question and study design alone, before any results exist.

What Science Actually Is: Kuhn and Lakatos

Thomas Kuhn shifted the focus from how science should work to how it actually does work in practice. In The Structure of Scientific Revolutions (1962), Kuhn examined what scientists across centuries actually did, and found that most science, most of the time, is what he called “normal science.” This is puzzle-solving within a paradigm, which is a shared framework of assumptions, methods, and exemplary problems whose fundamentals are not up for debate. “Mopping-up operations,” Kuhn wrote, “are what engage most scientists throughout their careers.” Anomalous results tend to be logged or deprioritized. Only when anomalies accumulate into a crisis does the broader framework itself get replaced.

If Kuhn is right, then Merton and Popper described a science that never quite existed. Scientists defer to prestige, and they work to protect their preferred theories. Given the incentives they face in the competition for tenure and scarce research grants, protecting their theories is often the rational strategy. But it means science as practiced is not science as prescribed.

Imre Lakatos accepted Kuhn’s descriptive point—scientists do rationally protect their core theories—while rejecting the conclusion that theory change therefore reduces to “mob psychology.” If no rational standard exists for judging one paradigm against another, then scientific revolutions are decided by whichever side wins more converts. Crowd dynamics, not evidence, wins the day.

His solution was to evaluate research programs rather than just individual theories. Every program, in his framework, has a protected core of commitments and a surrounding belt of auxiliary hypotheses that absorb anomalies. All programs accumulate anomalies; what distinguishes them is their direction. Progressive research programs generate novel predictions that pan out. Degenerating ones merely apply patches—like a company that revises its earnings guidance downward every quarter while insisting the underlying strategy is sound. Metascience at its best is a Lakatosian exercise, asking whether a field is progressing by making new predictions that hold up, or degenerating, by patching existing theories in response to every new anomaly.

Polanyi and the Limits of Mandating Rigor

That leaves Michael Polanyi. A world-class chemist before he turned to philosophy, Polanyi argued that scientific knowledge is largely tacit. Researchers know more than they can articulate, which means their knowledge cannot be fully audited or proceduralized. In “The Republic of Science” (1962), he described science as a spontaneous order—a system, like a market, in which thousands of researchers each adjust to what the others have found, producing a coordination that no central planner could design or replicate. “Any attempt at guiding scientific research towards a purpose other than its own is an attempt to deflect it from the advancement of science,” he warned.

That warning bears directly on the present moment. Metascience began as an insurgency of bloggers, responding to failed replications and promoting open-source tools. It is now a research program of governments. Economists can predict what will likely happen next. Goodhart’s law holds that when a measure becomes a target, it ceases to be a good measure. Journals learned to game impact factors, academics learned how to game citation counts, and there is no reason to think preregistration or replication are immune from these forces. The tacit dimension of science is precisely the part that top-down reform cannot reach, which risks neutering this latest attempt at reform.

None of this is exactly an argument against the metascience agenda, most of which constitutes an improvement on the status quo. However, a fair critique of the movement is that its agenda is too top-down. A federal mandate risks freezing one version of best practices into place and applying it uniformly across fields as different as particle physics and social psychology. These are fields that may not even mean the same thing by “replication.” The lesson of Polanyi is that the rules work best when they emerge from the practitioners who bear their costs. A mandate can make every researcher fill out the preregistration form, but it cannot make the form mean anything.

Keep Testing — Including the Tests

Merton and Popper taught us what science should aspire to be. Kuhn showed us what it is. Lakatos built standards for judging the differences between the two. Polanyi warned against choking off science by command. Modern metascience is an attempt to bring science as practiced closer to science as prescribed. The old ideals are being run as testable hypotheses against Kuhnian reality. But the Polanyian question of who should do the fixing, at least for now, is sitting unresolved.

The only credible answer is to apply metascience’s own method to metascience itself. Treat the reforms as conjectures. Do preregistered studies replicate at higher rates? Do data-sharing mandates produce usable data, or just more academic theater? Does paying for replication buy knowledge, or bounty-hunting? Nobody knows yet, and that is the point. Science is the process by which we decide, collectively, what is true. The one norm that would honor all these great thinkers is to keep testing. Including the tests.

* * *

Further reading: Robert K. Merton, “The Normative Structure of Science” (1942); Karl Popper, Conjectures and Refutations (1963); Thomas Kuhn, The Structure of Scientific Revolutions (1962); Imre Lakatos, “Falsification and the Methodology of Scientific Research Programmes” (1970); Michael Polanyi, “The Republic of Science” (1962).

The White House Office of Science and Technology Policy has released a new science strategy. Timed to coincide with the nation’s 250th anniversary, the document, titled “Science: A New Golden Age,” openly casts itself as a successor to Science: The Endless Frontier, a famous 1945 report that convinced Washington to fund basic research and which set the stage for the National Science Foundation. While that is a lot to claim, the new report comes with teeth, OSTP and the White House budget office have already directed agencies to start implementing its recommendations in fiscal year 2028.

There is a great deal to like here. The report’s account of institutional sclerosis is unusually frank for an official document, and its appetite for experimentation with new funding mechanisms appears sincere. Yet for all its market friendly language, the report remains wedded to a model in which government experts decide which projects deserve money. A bolder approach would let investment returns do more of the deciding, and the administration already possesses the tools to make that happen.

What the Report Recommends

The recommendations in the report span four chapters, and they read as four big bets, to borrow a word the report itself favors.

The first bet is on individual scientists rather than legacy institutions. The animating idea is that talent deserves the funding, not the university that houses it. Fellowships would travel with the graduate student who earned them. Young researchers would get backing to run their own programs earlier in their careers, and the boldest would receive grants with horizons long enough to pursue something that might fail for a decade before it works. Consensus peer review, which tends to remove the strangeness off ambitious proposals, would loosen its grip. A golden ticket system could let one convinced reviewer fund an idea the rest of the panel rejects, or a prize would pay only when someone actually delivers.

The second bet is on new kinds of research organizations, born of the observation that many important problems fit nowhere in the current landscape. They are too big for an academic lab, too interdisciplinary for a single department, or too far from market for a private company, so they fall through the cracks between institutions. The report wants to fill those cracks with a wider cast of performers. Consider the newly launched X-Labs, an NSF program that funds research organizations outside universities. It assembles teams of professional scientists for a few years to break a specific bottleneck and then disbands them. New agencies in the mold of DARPA would hand individual program managers real authority to make bold bets. And a handful of institutes would do the opposite, giving top researchers stable funding to chase fundamental questions for as long as the questions demand.

The third bet is on clearing the ground for builders. Here the report reads like a deregulatory manifesto. It champions permissionless innovation and regulatory sandboxes, would speed up permitting, and would open the national laboratories to startups that could never afford such facilities on their own. It calls for grand challenges in the tradition of the Apollo Program and the Human Genome Project. A companion chapter turns to manufacturing and the skilled trades, arguing that the tacit knowledge carried in the hands of machinists and technicians matters as much to technological leadership as anything published in a journal.

The fourth bet is on artificial intelligence. The report wants to fully fund and expand the Genesis Mission, the AI for science initiative President Trump launched by executive order in November 2025, and frames its goal as nothing less than doubling the productivity and impact of American science within a decade. Because AI trained on a flawed scientific literature will simply regurgitate error, it pairs that ambition with Gold Standard Science requirements for reproducibility and data sharing.

Government as Capital Allocator

Taken on its own terms, much of this agenda is sensible, and some of it is overdue. The candor in the report about current administrative bloat is refreshing. Several proposals could do real good. Still, a tension runs through the document. The report repeatedly instructs agencies to behave like investors. “Federal research agencies should evaluate their own performance as capital allocators,” it says, and should “construct portfolios with the intentionality of a serious investor.”

The aspiration is admirable, but the analogy breaks down at precisely the point that matters most. A venture capital fund that picks losers loses its capital and, eventually, its limited partners. A federal agency that picks losers rarely faces any financial penalty, and it may even receive a larger appropriation, since failure is often read as evidence that the problem is hard and requires more money. No golden ticket, fast grant, or metascience unit changes that feedback loop. Money still flows from experts and program managers to the projects they favor or that align with the political or moral priorities of the moment.

The report asks agencies to think like investors while leaving them structurally incapable of investing. That is the sense in which the document, for all its reformist energy, remains a plan for smarter bureaucrats rather than a blueprint for learning what works and building the incentives and feedback loops that let innovation flourish.

The Sovereign Wealth Fund Opportunity

Here the administration is sitting on an unused tool of its own making. In February 2025, President Trump signed an executive order directing the development of a plan for a United States sovereign wealth fund. Since then, the government has taken direct equity stakes in strategic companies, including a stake of just under 10 percent in Intel, and a roughly 15 percent position in MP Materials, the rare earth producer. Skeptics warn that government shareholding carries dangers of its own. But the Intel deal was well structured to avoid political interference, as it included a passive position with no board seat and no governance rights.

The federal government has shown it is willing to invest as an owner in firms at the cutting edge of technology and central to the national interest. The science agenda and the wealth fund agenda belong together. Where research has plausible commercial application, the government could take equity positions in research driven ventures instead of writing grants, ideally as a passive minority investor alongside private capital. Returns would flow back to taxpayers and could be recycled into the next generation of projects. More important, returns would supply the signal that grantmaking lacks. Projects that create value should attract more capital. Projects that fail to create value would shrink.

This model has an impressive track record abroad, though not an automatic one. State investment vehicles have produced disasters as well as triumphs, from the looting of Malaysia’s 1MDB sovereign wealth fund to money losing Washington bets like Solyndra, but what separates the successes from the failures is governance design rather than luck. The funds that work require private capital to invest alongside the state, keep politicians out of individual investment decisions, and build in exit options.

When Morris Chang founded TSMC in 1987, Taiwan’s government supplied roughly 48 percent of the founding capital through its National Development Fund after private investors hesitated. The fund still holds about 6 percent of what is now one of the most valuable companies on earth, a stake worth vastly more than the original outlay. Israel’s Yozma program committed $100 million of public money in 1993 to seed ten venture funds, required each to bring in an experienced foreign venture partner, and gave private investors an option to buy out the government’s share. Annual Israeli venture fundraising grew from about $40 million in 1991 to $2.7 billion by 2000, and the program privatized itself within five years. Singapore’s Temasek, which holds the state’s stakes in companies, reported a net portfolio value of S$518 billion in its July 2026 annual review and calls its artificial intelligence strategy integral to how it invests, deploying capital across the AI value chain from data centers to foundation models. Norway’s fund, built from oil revenue and now worth around $2 trillion, demonstrates that a democratic government can manage a vast public portfolio with transparency and discipline.

None of this means that every project must pay off immediately. Some of the most valuable projects pay off on timescales no investor will wait out. The particle accelerators and early genomic research of the last century generated no revenue for decades, then underwrote entire industries. Where returns arrive that slowly, they cannot guide funding decisions in real time, and expert judgment still helps fill that gap. And one big bet can end up paying off enough to fund ten losing gambles.

An Improvement That Could Go Further

Science: A New Golden Age is a better blueprint than the status quo it seeks to replace. Its diagnosis is accurate and its experimental spirit is welcome. But the report walks up to its own best idea and stops. It borrows the language of investment while preserving the machinery of allocation by committee. If the administration were to marry its science agenda to its sovereign wealth fund ambitions, letting realized returns rather than expert panels direct a growing share of public research capital, the golden age it promises would rest on much firmer foundations.

A new NBER working paper by Caitlin Myers and Ezekiel Hooper asks whether the arrival of the iPhone helped lower fertility in the United States. Myers and Hooper estimate that the iPhone explains a substantial share of the fertility decline after 2007.

The authors exploit a quirk of the early smartphone market. From June 2007 through February 2011, the iPhone was available in the United States only through AT&T. A county with broad AT&T mobile broadband coverage was therefore more exposed to the early iPhone than a county with little AT&T coverage. The paper uses county-level birth data from 2003 through 2011. Since the iPhone launched in June 2007 and births occur months after conception, the authors treat 2008 as the first birth year that could plausibly reflect exposure to the new device.

The estimated effects are largest for young women. The paper reports reductions in births of 4.5 to 8.0 percent for women ages 15 to 19 and 3.2 to 6.6 percent for women ages 20 to 24. Effects at older ages are smaller. Scaling the county estimates to the national level, the authors conclude that the iPhone accounts for 33 to 52 percent of the fall in the general fertility rate between 2007 and 2011. The mechanisms they discuss include reduced in-person interaction and more digital sexual substitution such as pornography, leading to less sex among young adults.

How this fits in the fertility literature

The broader literature on fertility decline points to a multi-causal explanation, rather than a single dominant cause. The long-run decline in fertility has been closely associated with falling child mortality, rising female education, urbanization, and reliable contraception. In rich countries today, work pressure, the opportunity cost of childrearing, housing costs, child care costs, economic uncertainty, delayed marriage, and changing norms about family size all have empirical support in some settings. Policies that lower the direct cost of children appear to raise births, but often only modestly and sometimes by shifting timing rather than completed family size.

The smartphone hypothesis belongs to the technology and culture branch of this literature. It is closest to work showing that media and communication technologies can change aspirations, time use, relationship formation, and childbearing decisions. That branch also includes experimental evidence on social media use, such as research on Facebook deactivation, which suggests that online platforms can alter offline socializing. The iPhone paper is best understood as a proposed explanation for the sharp decline in U.S. fertility after 2007, particularly among younger adults and for births that may have been unintended.

The paper’s methods

Myers and Hooper estimate the iPhone’s effect using two complementary research designs. The two methods make different assumptions, but the results point in the same direction.

The first method is an entropy balanced Poisson event study. In plain English, this approach compares births in high AT&T coverage counties with births in low AT&T coverage counties, while reweighting the control counties so they look more like the treated counties on observable characteristics like racial composition or political orientation. The key feature of this design is that it largely takes the observed pre-2007 pattern as given and asks whether birth rates diverged after the iPhone became available. That makes the design relatively transparent, but it also means that the estimate could be biased if high-AT&T counties were already experiencing faster fertility decline for reasons unrelated to the iPhone.

The second method is synthetic difference in differences, or SDID. This approach builds a comparison group in two ways. First, it gives more weight to low-AT&T counties whose pre-iPhone fertility patterns look most like those of high-AT&T counties. Second, it gives more weight to the pre-iPhone years whose fertility patterns are most similar to those in the treated counties just before the iPhone's introduction.

The goal is to build a more accurate estimate of what would have happened after 2007 in the absence of the iPhone. That creates a risk, however. Because SDID reweights both the control counties and the pre-iPhone years, it can reduce visible differences between the two groups before 2007. That is useful if those differences reflect noise. But it is a problem if they reflect real differences in fertility trends.

In the paper’s results, the entropy balanced Poisson estimates tend to be smaller, while the SDID estimates tend to be larger (see figure). The authors also report several robustness checks.

Potential Concerns

A recent review of synthetic control methods by Yiqing Xu is useful for judging the paper. Xu argues that the field has largely converged on a few practical lessons. Most important for this paper, Xu describes SDID as one of the strongest descendants of classical synthetic control. That assessment bodes well for Myers and Hooper, who do not rely on the more restrictive original form. That said, there are several reasons to read the estimates with caution.

The short pre-period problem

The first concern is the length of the pre-treatment period. The county panel begins in 2003 and treatment begins in 2008, which leaves only 2003 through 2007 to learn how treated and control counties moved together before the iPhone. Xu warns that synthetic control methods are very demanding in this dimension. With only a small number of pre-treatment observations, the weights can fit noise, and there are few years left for independent validation. This paper has fewer than the roughly ten pre treatment periods that Xu already treats as potentially thin.

A short pre period can make a comparison group look better than it is, especially when the treated and donor counties differ systematically. High AT&T coverage counties were more urban and more affluent than low coverage counties. The authors adjust for those differences, but adjustments can be imperfect.

The trend instability problem

A second concern is how the two methods handle pre-2007 fertility trends. Fertility was not changing uniformly before the iPhone arrived. It was moving differently across age groups, between urban and rural areas, among married and unmarried women, and across generations. That matters because high-coverage and low-coverage counties may have had different mixes of people and places whose fertility was already on different paths. SDID treats the pre-treatment movement in the data as information that helps construct the no-iPhone counterfactual, while the entropy balanced Poisson approach largely takes the pre-treatment path as given.

That creates the central interpretive question. If the pre-2007 drift was unrelated to the iPhone and had mostly run its course before the device arrived, the smaller Poisson estimates may be cleaner. If it reflected county-level dynamics that would have continued even without the iPhone, SDID may be closer to the truth. The paper cannot fully distinguish between those interpretations.

What to make of the result

My overall assessment is that the study is methodologically defensible, but the evidence remains suggestive rather than conclusive. The paper presents evidence that early iPhone access may have reduced births among younger women. However, the much larger claim that the iPhone explains one third to one half of the overall fertility decline from 2007 to 2011 requires a very large social effect from a technology whose adoption was still limited in its early years. Such a large effect would likely require the iPhone to influence not only its users, but also the broader social environment in which relationships and childbearing decisions occur. The safest reading is that smartphones now belong in the fertility decline literature, especially as a possible explanation for the post-2007 decline in young and unintended births. This do not replace the better-established evidence that declining fertility reflects deeper economic and social changes that long predate the smartphone era.

Edmund Phelps died on May 15, leaving behind one of the more inquisitive and humane bodies of work in modern economics. He was ninety two. Phelps was awarded the 2006 Nobel Prize in Economic Sciences for his work on the tradeoffs at the heart of macroeconomic policy. His best known contributions concerned inflation and unemployment, especially the way expectations shape both. He helped bury the comfortable old view that policy makers could purchase permanently lower unemployment with a little more inflation.

Yet Phelps’s thought reached deeper than the Phillips curve, the once influential idea that inflation and unemployment move in opposite directions. A recurring concern in his work was how economic choices affect welfare across time. This was true in his work on inflation. It was also true in his work on growth. A question that seemed to interest him was what today’s choices make possible for tomorrow.

The Golden Rule of Accumulation

That concern may have led him early in his career to the Golden Rule of accumulation. Phelps argued that a society guided by equal concern for generations should aim for the capital stock that maximizes sustainable consumption. In a 1961 essay in the American Economic Review, Phelps framed the issue through a fable about a kingdom called Solovia. His setup was deliberately simple. Labor grew. Capital accumulated. Output rose. And a fixed share of income was saved. The question was which saving share would give each generation the highest sustainable consumption along a continuing path of growth.

His answer was the Golden Rule. In Phelps’s simple competitive model, the ideal saving rate matched capital’s share of national income. In the more familiar classroom version, the economy should keep accumulating capital until the net return on capital equals the economy’s growth rate. This achieved the level of saving that maximized steady state consumption across generations. Phelps’s contribution was to turn an ethical intuition about each generation doing unto others as it would have done for itself into a clear economic benchmark.

Imagine an economy with very little capital. Workers lack the tools and machines that would make them more productive. In that world, saving more can raise future consumption because each new investment adds a lot. As the economy grows richer, however, the gains from additional capital gradually diminish. The Golden Rule aimed to identify the level of saving at which steady state consumption is highest. A steady state is a situation where the basic ratios in the economy stop changing over time, even though the economy itself may still be growing.

Predecessors and the Nobel Verdict

The Golden Rule was not created in a vacuum. Frank Ramsey had already posed the basic question in 1928, asking how much of a nation’s income should be saved rather than consumed. Robert Solow later supplied the growth model that made it possible to analyze saving, capital accumulation, and balanced growth with new clarity. Other economists reached related results. Phelps’s contribution was to make the benchmark memorable, teachable, and morally transparent. The Nobel committee later wrote that his treatment shaped the subsequent literature more decisively than any of the earlier attempts.

Phelps’s fable also had limits, however. The 1961 paper compared long run, steady state positions. It did not settle the hard problem of moving from one saving path to another. If a country needs to save more to reach the Golden Rule, the current generation must consume less while later generations gain. That is a genuine political problem, which the economics alone cannot resolve.

A Society Can Save Too Much

Phelps returned to the issue in a 1965 sequel. There he sharpened one of the most arresting implications of the Golden Rule. A society can save too much. If the capital stock is too large, cutting saving can raise consumption now and in the long run. Imagine an economy already rich with capital. More and more of society’s saving goes toward sustaining a large capital stock as the economy grows. At some point, the economy devotes so many resources to accumulation that households could consume more by saving less. The Golden Rule aimed to identify that tipping point.

Later economists would place this idea under the heading of “dynamic inefficiency,” but Phelps had already exposed the heart of the matter. Thrift may be a virtue, but Phelps showed it is not an absolute good, and consumption is not automatically selfish. Growth policy can fail through excess thrift as surely as through indulgence. The right balance depends on returns, growth, and population. Phelps supplied a framework for thinking about that balance.

Overlapping Generations and Discounted Utility

Peter Diamond’s overlapping generations model extended the Golden Rule idea to an economy where people are born at different times. In the model, young people save for retirement while older people spend down assets. The choices of one generation affect the opportunities of the next, but no generation can trade directly with all the others. As a result, voluntary market decisions tend not to produce the Golden Rule level of capital. The economy can end up with too much or too little savings. Diamond showed that when an economy has too much capital, government debt or pay as you go pensions can transfer resources from the young to the old, reducing excessive saving and raising consumption for multiple generations. Whether real-life public pension programs work that way in practice remains to be seen, but the broader lesson is markets can work well within a generation and still fail to allocate resources efficiently across generations.

David Cass and Tjalling Koopmans pushed growth theory in a different direction. Rather than ask which saving rate would maximize sustainable consumption for each generation, they asked how a planner would choose consumption and saving over time if society exhibits impatience. Their modified Golden Rule generally selects a lower capital stock than Phelps’s equal generation benchmark. The Cass-Koopmans approach of discounting future utility implies that later people’s consumption and welfare count for less in present decisions, while Phelps’s unmodified rule asks what follows if generations are treated symmetrically.

Extending the Golden Rule to Public Finance

Phelps’s framework still matters today because governments constantly make decisions that affect how much society saves and invests. Public budgets, infrastructure spending, entitlement promises, and education systems all shift resources across generations. The central question is which generation pays and which generation benefits.

The Golden Rule cannot tell Congress which programs deserve funding, nor does it resolve disputes about inequality or redistribution. But it does offer a framework for imposing discipline on public finance. Policies that raise future productive capacity stand on different ground from policies that simply move consumption forward in time.

If an economy saves too little, productive public investment can leave future generations better off. But when government spending simply finances current consumption without increasing future output, later generations bear the cost, on top of any sacrifice that borrowing requires today.

Capital Broadly Understood

Phelps did not treat capital as only machines, buildings, and equipment. He also applied Golden Rule reasoning to research, education, and the spread of knowledge. He saw that research and education also require sacrifice today in exchange for greater productive capacity tomorrow. Money spent on science may reduce current consumption, but it can raise future output by creating new knowledge. Money spent on education can do the same by helping workers learn better methods. In that sense, research and education function like capital. They are investments in future capability.

This connects Phelps’s early growth theory to the themes that later defined much of his work. He wrote about innovation, mass flourishing, and the dignity people find in creative work. The Golden Rule belongs to that same moral universe. Economic growth was never just a matter of piling up capital. It is about building the conditions under which people could do more, create more, and live better.

By invoking the Golden Rule, Phelps made clear that capital accumulation was both an economic problem and an ethical one. Each generation stands between what it inherited and what it passes on. It cannot consult the dead, and the unborn cannot vote. A rule of reciprocity is a way of bringing those absent generations into the calculation. There is humility in Phelps’s framing. He did not claim that a planner can know everything. He used a model to clarify a tradeoff that democracies often obscure, stripping away noise while preserving responsibility.

Later in life Phelps became associated with promoting mass flourishing and the creative powers of ordinary people in open societies. That work focused less on formal growth models and more on the conditions that allow people to find meaning in economic life. Still, the connection to the Golden Rule is close. Both parts of his work treated growth as a way of expanding human possibilities.

What the Rule Still Teaches

Phelps’s death invites the usual tributes to a Nobel laureate. They are deserved. But the more useful tribute is to recover the question he put before economics at the start of his career, namely how much a society should consume, how much it should invest, and what obligations the present generation owes to those not yet here to speak.

The Golden Rule does not make political decisions any easier. It does, however, force societies to confront the tradeoffs they would often prefer to ignore. More than six decades after Phelps wrote his fable, his ideas contain important lessons for economic statesmanship. A prosperous society should understand why it is saving, who bears the cost, and who stands to benefit.

Venezuela has captured the world’s attention with the dramatic January capture of its president, Nicolás Maduro, in a U.S.-led operation. Argentina, meanwhile, has become another focal point in the region, as Javier Milei’s government pushes an aggressive program of market-oriented reform.

While that attention is warranted, elsewhere in Latin America a quieter political and economic shift may be taking place in Bolivia. Long overshadowed by its more dramatic neighbors, Bolivia now stands at a potentially pivotal moment. If a new government follows through on promised reforms, the country could emerge as an important economic success story in the region.

A Historic Change of Power

In August of last year, Bolivians went to the polls for the first time without the Movimiento al Socialismo (MAS) as the dominant force organizing their politics. After multiple rounds of voting, culminating in an October 19 runoff, Rodrigo Paz Pereira, a senator and leader of the Christian Democratic Party (PDC), won the presidency with 55 percent of the vote. He was sworn into office on November 8th.

Paz’s victory ended nearly two decades of MAS dominance, a period spanning Evo Morales’s presidency (2006–2019) and Luis Arce’s administration (2020–2025), during which Bolivia’s economy became one of the most state-directed in the hemisphere. Paz, a pro-market conservative, inherited an economy in crisis, one marked by severe dollar shortages, chronic fuel deficits, and the highest inflation the country had seen in four decades.

Bolivia’s election was also historic in a structural sense. It was the first time the country’s presidency was decided by a runoff vote, signaling a genuine pluralization of its political landscape after years of single-party dominance.

How Bolivia Makes Law: A Civil Law System Without a Regulatory Code

Understanding what the Paz administration can actually do and how fast requires an appreciation of Bolivia’s legal architecture. Bolivia operates under a civil law system, descended from the Spanish Napoleonic tradition. Unlike common law systems like in the United States, which develop much of their law through court precedent, Bolivia’s legal framework is established primarily through codified statutes and executive decrees. Bolivia has civil, commercial, criminal, and procedural codes, as well as a landmark 2009 Political Constitution.

Laws passed by the Plurinational Legislative Assembly, once signed by the president, must be published in the Gaceta Oficial del Estado Plurinacional de Bolivia (the Official State Gazette) to take legal effect. Under the 2009 Constitution, a law becomes mandatory from the moment of its publication, unless the law itself specifies a different effective date. The Gaceta is the sole authoritative channel through which legislation is promulgated and made binding.

Bolivia does not have a standalone regulatory code comparable to the U.S. Code of Federal Regulations, where all executive agency rules are organized, indexed, and made accessible in a unified format. Rules, regulations, supreme decrees, ministerial resolutions, and other sub-legislative instruments are published as individual Gaceta entries, scattered across hundreds of editions with no consolidated index comparable to an administrative code. Finding and understanding the full scope of regulation in any given sector requires locating and cross-referencing a large number of individual official gazette publications.

American states have faced analogous challenges. Arkansas, for instance, long lacked a consolidated regulatory code. It was only through Act 64 of 2021, which directed the codification of all state administrative rules into a unified Code of Arkansas Rules, that the state took meaningful steps toward creating the kind of consolidated, regulatory code that transparency and accountability require. Bolivia, in 2026, is in roughly the position Arkansas was before that reform. It is a state with an official gazette, active executive branch activity, and no consolidated regulatory code to make the totality of the regulatory burden visible or manageable.

The President’s Decree Power: Vast and Direct

A defining feature of Bolivia’s regulatory system, and a major source of regulatory risk, is the president’s authority to issue Supreme Decrees (Decretos Supremos).

Under Article 172 of the 2009 Constitution, the president has explicit power to “issue supreme decrees and resolutions.” Ministers of State may propose drafts of supreme decrees and sign them jointly with the president, but the authority is ultimately executive. Although supreme decrees may superficially resemble executive orders in the United States, the comparison is misleading. U.S. executive orders generally operate as directives to officials within the executive branch, whereas Bolivian supreme decrees can more directly and broadly affect the regulatory environment, the public, and the economy.

Unlike the rulemaking process in the United States where agencies must typically publish proposed rules, accept public comment, and justify their final decisions with a record, Bolivia’s supreme decree process imposes far more limited constraints in its procedural architecture. A decree is drafted, approved by the Council of Ministers if warranted, and published in the Gaceta Oficial. It takes effect immediately.

This mechanism has enabled dramatic and rapid policy change throughout Bolivia’s modern history. It has also produced many consequential and controversial regulatory interventions in recent Latin American memory.

A Gallery of Controversial Decrees

In many respects, the history of Bolivia’s supreme decrees is the history of Bolivia’s economy. This is best illustrated by some of the more controversial decrees issued in recent years.

Supreme Decree 21060 (August 29, 1985): Shock Therapy

Perhaps no supreme decree in Latin American history was more consequential than Supreme Decree 21060, issued by President Víctor Paz Estenssoro on August 29, 1985. Bolivia at the time was experiencing one of the worst hyperinflationary episodes in recorded economic history, with annual inflation exceeding 20,000 percent. The decree was the Bolivian government’s radical response.

In a sweeping document spanning well over a hundred articles, DS 21060 implemented the so-called “New Economic Policy” in a single stroke. It established a unified, flexible exchange rate; liberalized import and export regimes; freed prices across the economy; dismantled a variety of state monopolies; decentralized the state oil company YPFB into subsidiaries; and restructured the Bolivian Mining Corporation (COMIBOL) along similar lines. It also eliminated a host of targeted tax exemptions, consolidated wages, and restricted public sector hiring. The decree was drafted with IMF support and is widely credited with ending Bolivia’s hyperinflation, though at significant social cost, including large-scale layoffs in the mining sector.

DS 21060 demonstrated what Bolivia’s supreme decree power can accomplish when used with market-oriented intent. It can enable sweeping structural change, implemented rapidly, without extended legislative negotiation.

Supreme Decree 28699 and 28701 (May 1, 2006): Renationalization and Labor Retrenchment

Twenty-one years later, Evo Morales used the same executive instrument to reverse course. On May 1, 2006, International Workers’ Day, his administration simultaneously issued two landmark decrees.

Supreme Decree 28699 reversed the labor market flexibility introduced by DS 21060. It prohibited dismissal without just cause and guaranteed employment stability for workers across the private and public sectors, hamstringing the ability of companies to adjust their workforces in response to economic conditions.

On the same day, Supreme Decree 28701, named the “Heroes of the Chaco” decree, nationalized Bolivia’s hydrocarbon sector. It transferred “absolute control” of all hydrocarbon production to the state-owned YPFB and required foreign energy companies to hand over production to the state and sign new contracts within 180 days or exit the country. The decree was a defining moment of the Morales era and triggered months of intense international negotiation with energy firms from Brazil, Spain, France, and elsewhere.

Both decrees illustrated the flip side of Bolivia’s concentrated executive power. The same mechanism that enabled rapid market liberalization in 1985 could just as easily enable rapid nationalization in 2006.

Supreme Decree 29894 (February 7, 2009): Restructuring the State

Supreme Decree 29894 was issued just days after Bolivia’s new plurinational constitution came into force. It restructured the entire executive branch of government, formally transitioning the country’s institutional architecture to the “Plurinational State of Bolivia.” It redefined ministry jurisdictions, established new state agencies, and set out the reorganized powers of the executive under the new constitutional framework. The decree illustrated how Bolivia’s leaders use supreme decrees not just for economic policy, but for constitutional implementation, bypassing or supplementing the legislative process even for fundamental institutional change.

Supreme Decree 1597 (June 5, 2013): Regulating Civil Society

Supreme Decree 1597, issued by the Morales government in 2013 as implementing regulations for Law 351, established the framework under which NGOs, foundations, and civil society organizations obtain, or lose, their legal status in Bolivia. The decree drew criticism from human rights observers for authorizing the revocation of an organization’s legal personality if it engages in activities beyond those specified in its founding statutes. Critics argued this provision created a bureaucratic mechanism for suppressing civic organizations that challenged government policy, and with some human rights organizations expressing concern about its potential chilling effects.

Supreme Decree 4436 (December 30, 2020): A Tax on Wealth

Supreme Decree 4436, issued by the incoming Arce administration in late 2020, established the Impuesto a las Grandes Fortunas (IGF)—a permanent progressive tax on individuals with net wealth exceeding 30 million Bolivianos (roughly $4.3 million at the time). Although formally implementing Law 1357, Supreme Decree 4436 was sweeping in effect, as it established the detailed rules, thresholds, and compliance framework that determined how broadly and forcefully the wealth tax would apply in practice. The tax was framed as a redistributive response to the economic damage wrought by COVID-19, but it attracted criticism from business groups who argued it would accelerate capital flight and worsen Bolivia’s already fragile investment climate.

Supreme Decree 5503 and 5516 (December 2025 / January 2026): The New Administration’s Trial by Fire

No episode better illustrates the double-edged nature of Bolivia’s decree-centered governance than the sequence of events that followed Rodrigo Paz’s inauguration. Inheriting an economy that had exhausted two decades of hydrocarbon revenues to sustain domestic fuel subsidies, Paz’s administration issued Supreme Decree 5503 on December 17, 2025, a comprehensive economic reform decree with over 120 articles. Its centerpiece was the elimination of fuel subsidies that had been in place for roughly twenty years, a fiscal adjustment that Paz framed as a necessary response to “20 years of looting and waste.”

The immediate effect was a surge in public transport costs and widespread protests, particularly in El Alto. Road blockades by transport workers and regional labor unions paralyzed parts of the country. Within weeks, Paz’s government issued Supreme Decree 5516 in January 2026, repealing some of the more contested provisions of DS 5503. The core commitment to market-based fuel pricing survived, but the administration’s willingness to use, and then partially walk back, a major reform decree within the span of a month demonstrated both the power and the fragility of Bolivia’s decree-based governance.

A subsequent decree, Supreme Decree 5517, declared a national “energetic and social emergency,” authorized exceptional fuel imports, and removed diesel from the list of controlled substances to facilitate private importation—another sweeping intervention achievable by executive action alone.

The Structural Lesson: Power Without Architecture

What the history of Bolivia’s supreme decrees reveals is a governance system of enormous executive flexibility and vanishingly little institutional constraint. When that flexibility is directed toward liberalization, as it was in 1985, the results can be transformative. When directed toward nationalization, redistribution, or the suppression of civil society, as in 2006 and 2013, the same instrument can do comparable damage.

Bolivia has a legal system capable of rapid change, but it lacks the checks and balances that make regulatory governance accountable. At a more practical level, it also lacks basic good governance features such as a consolidated regulatory code, a public comment process, regulatory review requirements, cost-benefit analysis for major rules, and periodic review of existing regulations for continued relevance.

This is the Arkansas problem, applied to a sovereign state. Without a unified regulatory code, the accumulation of supreme decrees and Gaceta entries creates a regulatory environment that is opaque, difficult to navigate, and resistant to systematic reform. Businesses, whether domestic or foreign, cannot easily determine what the full set of applicable rules is in any given sector. Policymakers cannot easily identify redundant or contradictory requirements. Reformers cannot easily target inefficiency. The regulatory state grows incrementally through decree, but never clears itself out.

The Venezuela Distraction and the Bolivia Opportunity

Venezuela and Argentina have consumed enormous attention from policymakers, journalists, and analysts focused on Latin American reform. That attention is warranted. The humanitarian dimensions of Venezuela’s crisis are severe, and its geopolitical implications are significant. Argentina, too, has drawn intense scrutiny, as President Javier Milei’s aggressive effort to dismantle long-standing controls, cut spending, and liberalize the economy has made the country a central test case for market-oriented reform in the region.

But Bolivia is a different story, and arguably one with more actionable near-term potential. The Paz administration has demonstrated both the willingness to use executive power for economic reform and, with the partial retreat on DS 5503, the pragmatic judgment to negotiate its pace. The country has a legitimate, democratically elected government with a market-oriented mandate. It has a legal system capable of implementing structural changes quickly.

What Bolivia lacks is the institutional infrastructure to make those changes durable and predictable. A consolidated regulatory code—one that organizes, indexes, and makes accessible the full body of executive decrees and ministerial rules—would be a foundational step. A formal process of regulatory review, analogous to what the United States conducts under executive orders governing regulatory planning, would be another. Methodologies for estimating the costs and benefits of major regulatory interventions could help the Bolivian government prioritize its reform agenda and demonstrate its fiscal seriousness to international investors and creditors.

None of this requires Bolivia to abandon its civil law tradition or its constitutional architecture. It requires only a decision to bring accountability and transparency to the regulatory instruments that already exist.

The 1985 liberalization reforms suggest that Bolivia, when it acts, can act decisively. History will not wait. Neither should Bolivia’s reformers, and neither should those in the international community who are watching for the next chapter of Latin American economic opening. All eyes may be on Venezuela and Argentina, but Bolivia is the story that shouldn’t be missed.

*This post was drafted with the assistance of AI tools. Any errors or omissions remain the responsibility of the author.

This week, the U.S. Supreme Court heard oral arguments in the case challenging the legal authority behind President Donald Trump’s tariffs. The press has highlighted that several justices appeared skeptical of the administration’s legal defense, prompting speculation that the tariffs could soon be struck down. But, as is often the case in Washington, much of the reaction is probably overblown.

Even if the Court were to rule against the administration, the tariffs would almost certainly return in some form under a different legal authority. Moreover, any subsequent legal challenges could take years to resolve and could easily run through the duration of Trump’s term in office. Moreover, the tariffs have been in flux since their inception, so even striking them down now doesn’t necessarily change where they are likely to end up.

Still, it’s worth reflecting on the debate surrounding Trump’s tariffs, because it has revealed as much about the state of economic discourse as about trade policy itself.

The Ineptitude of the Critics

What’s been most striking about the tariff debate is probably the ineptitude of many of Trump’s critics. This is remarkable, because there are ample reasons to oppose the tariffs. Yet the opposition has been defined mostly by hysteria, rather than by insight.

Many of the critics’ dire predictions simply haven’t come to pass. The economic sky never fell. The stock market has held steady. Inflationary pressures, to the extent they existed, were hardly the existential threat foretold. Instead of measured analysis, the public has been treated to alarmist commentary that betrays the pundit class’s shallow understanding of basic economics.

A central problem is that too many critics rely on static, short-run supply-and-demand logic. They present simple graphs as if they settle the matter, assuming that any distortion to short-run equilibrium must be bad. But such reasoning misses the broader, dynamic effects that policies like tariffs can have. By focusing narrowly on immediate changes in prices and output, the critics also ignore how trade policy sets broader national strategies and priorities.

The obsession with short-term data in the form of quarterly growth rates, consumer prices, and job counts also gives the illusion of scientific rigor but adds little to our fundamental understanding. From the public’s perspective, economists have cried wolf once too often. When the predicted disasters failed to materialize, it’s little wonder that many ordinary citizens conclude that the experts don’t know what they’re talking about. In this case, it feels like another victory for the wisdom of the common man over the technocrat.

A Sound Impulse Behind the Policy

None of this is to say the tariffs have been implemented perfectly. They’ve been inconsistent, at times arbitrary, and clearly shaped by political as much as economic motives. Yet, the underlying impulse to pursue a more deliberate, strategic industrial policy is undoubtedly a sound one.

For decades, U.S. trade policy has been guided by an almost religious commitment to globalization. Trump’s tariffs represented a break from that orthodoxy. Whatever their flaws, they signaled a willingness to think differently about national economic priorities. Most Americans, I believe, favor this shift toward a more intentional industrial strategy, even if they sometimes disagree on the details.

So while the tariffs have generated controversy in academic and policy circles, they’ve arguably moved the country in a more constructive direction. It’s one reason why Trump’s second term, at least thus far, has been one of the least controversial in recent memory. You can’t please everyone, of course. But if the new direction of policy makes some people uncomfortable, that may simply be their problem.

The familiar supply and demand diagram is the bread and butter of the economics discipline, and for good reason. It provides immediate, intuitive insights into the workings of markets and the predictable consequences of interference. Impose a rent control policy, and you’ll probably get a shortage of apartments for rent. Establish a minimum wage, and you’ll likely see reduced hiring, or, at least, some reduction of benefits to workers. These are reliable, almost mechanical results. In this sense, price theory is indispensable.

But while supply and demand analysis is immensely useful for thinking about the world at a point in time, it loses much of its power as a tool for making normative policy recommendations as events unfold over time. A particularly astute comment I once saw on social media noted that standard supply and demand analysis lacks a time dimension.

Many economists were quick to counter that their tools can handle time. One can do “comparative statics” that compare two points in time, for example. Or, one can speak of “long-run” equilibria where supply and demand elasticities change. Others pointed out that the quantity axis of the graph typically measures a rate: output per period. These are all true statements, but none of them refute the original point. Equilibrium is, by definition, a static concept. It describes an outcome market forces gravitate toward at a particular moment in time. However, that “ideal” becomes much less compelling once our frame of reference incorporates the element of time more fully.

Time and the Allocation of Resources

The evolution of welfare through time depends on how individuals use their incomes, in particular whether they choose to consume or invest them. This savings dimension is central to how economies actually evolve. Consider a protective tariff. In the short run, the tariff forces consumers to pay higher prices. Households have less purchasing power to spend on other goods and services, introducing a deadweight loss.

But that’s not the end of the story. If the tariff redirects spending away from cheap, imported consumption goods and toward domestic industries that produce capital goods, the composition of output changes, both at home and abroad. Over time, that shift toward investment might lead to greater productive capacity and higher wages, even if short-run consumption suffers.

I’m not arguing that tariffs are always or even usually justified. The point is simply that static analysis misses this intertemporal possibility. Many of today’s proponents of tariffs are quite open about their willingness to accept short-term pain for longer-term restructuring. When economists point to short-run losses as dispositive evidence against such policies, they are missing the point.

Beyond “Total Surplus”

Supply and demand analysis rests on welfare measures in the form of consumer and producer surplus. Economists typically assume these surpluses are comparable and additive. But in reality, surpluses are not homogeneous. There’s no guarantee that maximizing “total surplus” today maximizes or even leads to greater “total surplus” tomorrow. The composition matters. A dollar of consumer surplus spent on consumption doesn’t have the same dynamic effects as a dollar of surplus reinvested in capital. The usual supply and demand analysis misses this.

Perhaps a more fruitful objective than maximizing total surplus would be maximizing investment surplus—that portion of total surplus directed toward investment and future production. Consuming more today does little to expand tomorrow’s production possibilities. Investing more, by contrast, does. This doesn’t mean consumption is bad, since our living standards depend on it. But there are tradeoffs involved when we consume, which too many economists overlook. Supply and demand analysis has contributed to a tendency to gloss over the important differences in how consumers choose to spend their incomes.

Respecting, But Not Worshiping, Price Theory

One of the most valuable aspects of supply and demand analysis is that it helps clarify the distinction between private and social costs and benefits. The model reminds us that the sum of individual gains may be maximized in ways that look very different from decentralized “private” market outcomes. This is true within the static, snapshot-in-time framework that most often shapes economists’ thinking, but the principle holds more generally across time as well.

Like any model, supply and demand is a simplification. In this case, it is a photograph, and not a film. Treating it as a universal measuring stick by which to judge every policy, rather than as a powerful but limited tool of predicting certain market outcomes, does a disservice both to economics and to the complexity of the world the model seeks to explain.

Much of what governments and private charities aim to accomplish lies outside of the marketplace. They try to reduce poverty, improve education, cure disease, or protect the environment. These are activities whose success is typically not easily reflected in market prices. That is not to say markets are totally absent. These organizations spend money, hire workers, and purchase goods and services, all of which can show up in GDP. But the outcomes they seek to achieve are rarely profit-related. They are spending money to “do good,” however that’s defined.

This is where grants come in. A grant is a transfer of funds from one entity, like the government or a foundation, to another, with the intent of achieving some social purpose. On the one hand, you can think of grants like any other market transaction. The donor is “buying” the recipient’s service or effort. In that sense, the service is sometimes complete the moment the money is transferred. There is often no enforceable expectation for a particular outcome, only a loose expectation that the recipient will try to do something useful. If the donor and recipient interact repeatedly, informal expectations might build up, but these are social ties rather than contractual ones.

In this narrow sense, giving a grant is like consuming a “service.” When I donate to a charity, I am buying a feeling of satisfaction or a sense of contribution. Once I’ve given the money, that feeling is “consumed.” If we view grants this way, they should simply count as consumption in GDP, and the story can end there.

But most people expect more. They want the grant to accomplish something. That’s where the trouble begins.

Two Big Problems with Grants

First, there is the principal–agent problem. The donor (the principal) wants to achieve some goal, but the recipient (the agent) has something different in mind. Once the money is transferred, the agent may not act in the donor’s interests. There’s often little accountability in the nonprofit sector, and oversight mechanisms can be weak.

Second, there are knowledge problems. The donor usually lacks the information needed to know what’s working or even what’s possible. The recipient may not know either. Consider a federal grant to improve local education outcomes. Did test scores rise because of the grant, or because of other changes, say, in demographics or due to a new technology? Often no one can tell.

In my experience, most government grant programs don’t work. The few that do often owe their success to luck more than design. Take DARPA, the Defense Advanced Research Projects Agency. It’s famous for funding breakthroughs like the internet and GPS, but this may have had more to do with its good fortune of happening upon well-positioned scientists at precisely the right moment than with any replicable grant-making process. For every such success, there are countless quiet failures we never hear about. And of course, “success” can also come through fraud. Some projects appear effective on paper but collapse under scrutiny or once the money runs out.

Strings Attached (Even When They’re Not)

One appeal of grants is supposed to be their flexibility. Grants can constitute “no strings attached” funding. But government grants almost always come with strings. Even when they don’t formally, they acquire them over time. As donors and recipients interact repeatedly, recipients learn how to please donors, to produce what recipients think donors want. This can create a culture of performance theater rather than one of genuine problem-solving.

Markets, for all their flaws, have built-in ways to align incentives with outcomes and to spread knowledge through feedback mechanisms. Wages paid to workers motivate them to act in the interests of the firm. Payments from customers reveal when and where value is being created. Share prices illuminate how well a firm is converting resources into satisfaction. Equity links the interests of a firm—its managers and workers—directly to those of its investors and other benefactors.

These linkages aren’t perfect, but they are good enough to produce the sustained economic growth we’ve seen over the past century. That’s why I believe governments should experiment far more with equity-based funding and begin winding down many grant programs.

From Grants to Equity

Governments have historically funded important research in basic science and technology, much of which goes on to enrich private actors. Taxpayers, however, rarely share in the gains to the same extent. If instead of giving grants, governments took equity stakes, they could capture some of the upside when funded innovations succeed.

This doesn’t mean bureaucrats should micromanage innovation. A hands-off investment approach is possible. But equity stakes would give governments better signals about which projects are working and which aren’t. This approach would also align recipients’ incentives with public benefit, because success would enrich taxpayers too.

Critics might worry this would turn bureaucrats into bean counters, restricted by rigid financial metrics. I see the opposite happening. Having reliable data is liberating. When decision-makers have solid information at their disposal about what’s working, they gain the freedom to pursue bold goals within realistic boundaries. They can take more intelligent risks.

Governments can add value by pushing the technological frontier outward where the market on its own would never go. That is what we should want government to do: to pursue high-risk, high-reward ventures that the private sector won’t touch. Advancing a new technology is far more worthwhile than tinkering around the edges of an old one, but it is also far riskier.

Abandoning grants for equity stakes doesn’t constrain government. It empowers it and channels its energies in productive directions. The grant model is like flying blind. The equity model is like flying with a compass and a radar. True, you’re still bound by the laws of physics, but at least you know which way you’re heading and whether anything is in the way.

Having sound data to guide public spending is not limiting, it’s liberating. It allows us to be deliberate about our objectives. Once we can measure what works, our goals can be as ambitious as we dare to make them.

For much of the 20th century, the Federal Reserve was a dominant force affecting America’s macroeconomic stability. The classic boom–bust business cycle was largely a story of monetary policy missteps. Loose money drove booms, while tight money drove busts. Economists like Milton Friedman and Anna Schwartz traced the Great Depression and many subsequent recessions to these swings in liquidity. The Fed was always reacting to economic data with a lag, tightening just as the economy slowed, easing just as it began to recover, creating the stop–go cycles that characterized much of the postwar period in the U.S.

Until 2008, this portrait described the world pretty well. The Fed controlled the monetary base, and through the money multiplier, increases in the monetary base translated into expansions or contractions in broader measures of money. It wasn’t a perfect relationship, but monetary policy was fairly predictable in its effects.

The Post-2008 World

Then came the Great Financial Crisis. In October 2008, the Fed began paying interest on reserves, effectively transforming bank reserves into interest-bearing assets comparable to short-term Treasury bills. This seemingly technical change had profound consequences.

By paying banks to hold reserves, the Fed ended the traditional separation between base money and other government liabilities. The velocity of base money (the rate at which it turns over in the economy) collapsed. The money multiplier, long a staple of macroeconomics textbooks, ceased to describe how the modern financial system works.

In earlier eras, the assets the Fed bought—typically Treasuries—were imperfect substitutes for money. When the Fed expanded its balance sheet, it truly altered liquidity conditions. But today, T-bills and reserves are nearly indistinguishable. Both are short-term, “risk-free” interest-bearing claims on the U.S. government. When the Fed swaps one for the other, little changes in the financial system’s overall liquidity composition. The result is that now money creation is more driven by endogenous factors—forces inside the banking system affecting lending—than it was before.

Banks use Treasury securities much like they use reserves, as near-money instruments for settlement and as high-quality collateral in wholesale funding markets. Moreover, the growth of stablecoins, which are digital tokens often backed by T-bills, has created a parallel monetary system where Treasury securities directly underpin private money creation.

In this new environment, the fiscal authority, meaning Congress and the Treasury, effectively sets the pace of money-like asset creation. The supply of T-bills, driven by government deficits, has become a key determinant of liquidity and inflation dynamics.

This helps explain why massive fiscal packages, such as the Biden administration’s 2021 American Rescue Plan, triggered such pronounced inflationary pressures. When the government spends hundreds of billions of dollars into an economy, the larger deficit floods the system with new money-like instruments, raising aggregate demand.

Fiscal and Monetary Theories

Modern Monetary Theory (MMT) has often been dismissed as fringe economics, but it offers a useful lens for understanding this new regime. MMT correctly emphasizes that the government can spend money into existence through deficits and can later drain it back via taxes. In a world where Treasuries are the basis for money, the fiscal deficit is indeed a primary source of monetary expansion.

That said, MMT’s critics also have a point. MMT proponents often overlook political economy realities, tax revenues are rarely saved, Congress is far less nimble than a central bank, and MMT policy prescriptions tend to be fiscally reckless. But as a descriptive theory of how modern government finance works, MMT aligns pretty closely with post-2008 realities.

If fiscal policy is now a key determinant of liquidity conditions, then the deficit will tell us something about future inflation. This change may make us uncomfortable. It places enormous macroeconomic responsibility in the hands of Congress, an institution that garners little trust among the public and is designed more for deliberation than rapid stabilization policy. Yet the idea is not as far-fetched as it might seem.

For example, Milton Friedman once mused that monetary management based on rules set by Congress might be superior to discretionary independent central banking. Another framework, known as the Fiscal Theory of the Price Level, formalizes the relationship between government debt and inflation. According to this view, the value of money (and by extension, the price level) is determined by the government’s intertemporal budget constraint rather than the central bank’s control of the money supply. In plain terms, inflation arises when markets doubt that future fiscal policy will generate enough surpluses to support the current trajectory of debt levels. This position comes with the view that the Fed is ultimately subservient to the fiscal position of the government.

Nothing here contradicts supply-side theories of economic growth or Real Business Cycle (RBC) theories. Supply shocks like energy price spikes or productivity changes can still play a role in generating business cycles. Often, supply and demand shocks coincide. For example, a surge in perceived risk can raise the risk premium on interest rates, reducing investment and hiring (an RBC channel), while simultaneously boosting demand for safe assets like T-bills and bank reserves (a monetary channel).

Recessions typically feature lower interest rates amid falling output, reflecting a flight to safety. Nevertheless, RBC-style real shocks are often present during recessions, highlighting how monetary liquidity dynamics are deeply intertwined with the real economy.

Fiscal Is the New Monetary Policy

We have entered a new macroeconomic era. The dividing line between money and debt has blurred. T-bills circulate like near-money, stablecoins turn them into digital cash, and the government deficit determines how much of this “money” exists.

None of this means the Fed is totally inert. It can offset fiscal policy by adjusting the rates it pays on reserves. However, its current role is one more focused on fine-tuning financial conditions than steering the macroeconomy. That represents a big change from the past.

Increasingly, it is the fiscal authority that controls the real levers of liquidity, inflation, and aggregate demand. In that sense, fiscal policy is the new monetary policy. The challenge for our leaders is to recognize this change and design institutions that can manage it responsibly.

Fiscal dominance doesn’t mean chaos, but it does mean accountability must migrate from the technocratic realm of central banking to the far messier world of politics. Whether that transition proves successful will define the next chapter of macroeconomic history.

Economists have long relied on the notion of a utility function to model human decision-making. In its modern form, utility is not a measure of pleasure or happiness. Instead, it is a formal device for representing preference orderings. If a person prefers A to B, then we assign higher utility to A. That’s it.

Despite its simplicity, the idea of a utility function has proven remarkably powerful. From this starting point, economists derive demand curves, predict choices under constraints, and construct models to explain how individual decisions interact to produce market outcomes. But the simplicity comes at a cost. The theory rests on several strong assumptions some of which may not hold in practice.

How Economists Model Preferences

In the standard model, preferences are expressed as a binary relation. An individual either prefers A to B, prefers B to A, or is indifferent between them. From this web of pairwise comparisons, economists impose additional structure. The most important of these are:

Completeness - Every pair of alternatives can be compared.

Transitivity - If A is preferred to B, and B to C, then A is preferred to C.

Together, these create a consistent ordering, or what economists sometimes call a “rational” preference relation. From there, a utility function u(x) can be constructed to represent the ordering numerically.

Rethinking the Assumptions

The assumptions of completeness and transitivity give utility functions mathematical clarity, but they may not reflect how real people behave. Interestingly, relaxing these assumptions opens up some new possibilities.

Relaxing completeness recognizes that some values may be fundamentally incommensurable. What common measure could weigh justice against happiness, or dignity against freedom? Relaxing transitivity allows for cyclical or inconsistent preferences, acknowledging what behavioral evidence has shown for decades, which is that people’s choices often depend on context, framing, and emotion.

Besides being realistic, dropping the transitivity requirement also offers a convenient path out of the formal constraints that bedevil social choice theory, like Arrow’s Impossibility Theorem. Once we stop demanding perfect coherence, collective decisions become less impossible.

From Individual Rationality to Institutional Rationality

If individuals are not perfectly rational in their preferences, the search for a perfectly rational society is probably misguided. Instead, we should look to institutions as systems to manage this human inconsistency.

Institutions in the forms of markets, constitutions, and law are rationalizing mechanisms. They structure choices, stabilize expectations, and can constrain inconsistency and chaos in preference aggregation.

Where individuals may be intransitive, institutions can enforce transitivity through rules. Where people may be impulsive, institutions can impose delay and deliberation. Where preferences are incomplete, institutions can fill the gaps with norms and precedent.

Rationality becomes a property of the rules and procedures that govern decision-making, not of the agents themselves. This is what is meant by institutional rationality.

Conclusion

Mainstream economics deserves credit for what it achieves with so little. A model built on pairwise comparisons and consistent preferences can explain market behavior remarkably well.

However, the human mind does not store a ranked list of every possible outcome. Preferences are frequently constructed in the moment. Choices are influenced by habits, framing, norms, and rules, not from an internal, timeless utility map. Our tastes are an extension of social structures that both inform and discipline them.

Recognizing the benefits of institutional rationality and the shortcomings of mainstream utility theory doesn’t necessarily mean abandoning formal modeling. It means acknowledging that in the end, a society’s rationality doesn’t arise from every citizen ranking all possible outcomes consistently. It arises from the institutions that make collective life navigable despite that impossibility.

Markets themselves are one such institution, one that often incentivizes the kind of rationality economists have picked up on in their models. But that doesn’t mean human beings are inherently rational or markets always achieve the ideal outcome. The challenge for economists, then, is to better understand the rules and constraints that make an incoherent species coherent.

For much of human history, civilization has advanced through what the economist Friedrich Hayek famously called “spontaneous order,” which is a process by which social institutions, norms, and markets evolve organically from countless individual choices, rather than being deliberately designed as occurs under central planning. This invisible hand of coordination has been a hallmark of our civilization’s progress.

But that era may be coming to an end.

The Rise of Algorithmic Civilization

Whether we like it or not, we are entering a world where human life is increasingly guided, and perhaps even determined, by algorithms. Artificial intelligence is beginning to influence, and will soon dominate, the systems that shape our lives, including what news we see, what jobs we are offered, whom we meet, and even how we think about the world.

Already recommendation engines quietly nudge our preferences, large language models shape the contours of public debate, and predictive analytics play a role in decisions about hiring, credit, and even justice. Demand itself, which is the very foundation of market coordination, is no longer purely natural or emergent. It is being shaped, often invisibly, by systems that optimize for engagement and control. The “spontaneous” social forces that once governed civilization are gradually being replaced by this deliberate algorithmic guidance.

This does not necessarily mean we are entering a dystopia, but it does mean we are entering an era of AI determination, as civilization is increasingly guided by intelligent technology rather than emergent human interactions.

The Alignment Challenge

A key question AI researchers have been asking in recent years is whether AI systems are aligned with human values. The AI alignment problem is often framed as a technical issue. But even if society resolves that technical problem, deeper questions arise related to who defines the AI system’s intentions and whose values are being aligned with whose.

Many people feel threatened by these changes, and understandably so. When civilization is guided by emergent processes, no one is fully in charge and everyone participates. When it is guided by algorithms, the locus of control shifts to AI developers and corporations. We risk creating a world that may have many benefits that accompany it but is without moral depth, one that is not humane.

Toward a Deliberate Civilization

I am largely agnostic as to whether this shift is inherently good or bad in itself. It will surely be a mix of both. The important point is that it is happening, and it cannot be undone once the proverbial genie is out of the bottle. The era of pure spontaneous order is ending. What matters now is how we respond.

If humanity is to retain any semblance of control over its destiny going forward, we must establish institutions that can counterbalance algorithmic determination. We should not aim to destroy superintelligent AI or resist it, which may be a futile endeavor, but to counterweight its force to keep civilization anchored in human purpose and intention.

The only entity with the scale and legitimacy to provide such a counterbalance is likely to be the state. This is not a call for all-encompassing government control. The state is notoriously inefficient and often fails where markets succeed. Moreover, the state itself will inevitably utilize and be influenced by AI. But in certain areas where its aims are clear and results can be tested objectively against reality, the state remains capable of acting with decisiveness and conviction.

The Case for State Capitalism

What might such institutions look like? One promising model is state capitalism, a system in which the government plays an active but commercially disciplined role in markets rather than merely regulating them.

Consider, for example, a sovereign wealth fund, which is a state-owned investment vehicle that manages public assets for long-term returns. Such funds enable governments to pursue strategic and measurable objectives, from stabilizing national finances to investing in frontier industries, all while maintaining market-based feedback through their portfolio performance. Similar promise can be found in development banks, which can finance large-scale infrastructure projects and strategic industries that private capital alone might underprovide.

These forms of direct government participation in markets combine public goals with market accountability, allowing the state to guide investment toward socially valuable ends without abandoning the efficiencies and adaptive capacities of capitalism itself.

The key is building incentive compatibility. When institutions are incentive-compatible, they channel self-interest toward collective benefit, harnessing local knowledge and feedback mechanisms to discipline action. This can take many forms, from structuring public sector compensation contracts to reward performance, to requiring public initiatives to generate commercial returns,

The state can deliberately shape innovation. For instance, strategic investment funds can channel capital into frontier technologies while ensuring that taxpayers share in the upside by taking equity stakes in government-backed ventures. Development banks, in turn, can finance infrastructure and industrial projects, while guided by clear performance metrics and credible repayment plans.

In each case, the objective is to complement markets, providing directionality and complementarity without suffocating the adaptive forces that make markets so powerful.

The Choice Before Us

The choice before us is profound. Do we accept a future of AI determination, where civilization evolves according to algorithmic imperatives beyond human comprehension or control? Or do we build institutions that allow for self-determination, so humanity may retain a say in the course of its own destiny?

We cannot return to the world of pure spontaneous order. But we can ensure that the deliberate civilization we are entering remains, at its core, a human one.

Economists love precision, but sometimes that precision rests on convenient fictions. One of the most pervasive in economics is the idea that there exists a single, representative consumer whose choices can stand in for everyone else’s. This assumption sits quietly behind nearly every measure of “real” spending or output that economists use, from cost-of-living indices like the Consumer Price Index to real GDP.

It’s a practical fiction, to be sure, but a fiction nonetheless.

The Representative Basket That Isn’t

When economists adjust for inflation, they rely on price indices like the Laspeyres or the Paasche index. Each is built around a notional “basket of goods”, which represents the bundle of items that a “typical” consumer buys. The trouble is, there is no typical consumer. Some people spend more on rent, others more on healthcare or childcare. Some households are vegetarian, others carnivorous.

The Laspeyres index fixes the base-period basket to what someone bought in the past and asks how much that same basket would cost today. The Paasche index does the opposite, asking how much today’s basket would have cost in the past. Sometimes economists combine the two approaches by “chaining” price indices, linking together Laspeyres- and Paasche-style measures over time. A chained index updates the reference basket regularly, using a mix of past and current prices to capture changes in consumption patterns and the tendency of the consumer to substitute toward cheaper goods when the prices of close substitutes rise.

Both approaches implicitly assume that one person’s pattern of consumption can serve as a fair benchmark for everyone. That assumption works cleanly if everyone has homothetic preferences, that is, if richer and poorer individuals spend their money in the same proportions, just scaled up or down. Under such conditions, it doesn’t matter whose basket we use, because everyone is just a magnified version of everyone else.

The problem is real people don’t behave that way. As income rises, our spending patterns change. The share spent on necessities like food declines, while spending on luxuries, education, and leisure increases.

Adding to the complications, the people behind these data sets aren’t constant. Different individuals enter and exit the sample over time, making it hard to disentangle price increases caused by inflation from those driven by shifting preferences or demographics. Even the same individuals don’t stand still. Their tastes evolve as fashions and technologies change, and what counts as a “necessity” today may have been a luxury yesterday. The very act of measuring inflation with these indices assumes a kind of stability in consumption choices that real life simply doesn’t provide.

So, baked into every inflation adjustment is an implicit assumption about whose life counts as normal. The “representative” basket, in practice, is a political and statistical compromise. It’s an average that smooths over the diversity of human needs and wants.

Searching for a Better Anchor

The problem isn’t that real-income adjustments are meaningless. It’s that they hide normative judgments beneath technical machinery. When we report real GDP, we take the nominal value of output and divide it by a price index. But that index embeds assumptions about the representative agent and the stability of its preferences. So when economists say “real GDP grew 2 percent,” they’re not describing an objective fact about the economy as a whole so much as an interpretation of what happened to a particular kind of consumer.

There’s a lot to like about nominal GDP, if only because it avoids this pretense. It is the sum of final market transactions, measured at the prices at which those transactions actually occurred. Because inflation erodes the value of money, however, nominal measures gradually lose their meaning when we try to compare economic values across time. So long as inflation exists, we will need to adjust for it, whether it’s the government revising national income statistics or businesses updating prices in contracts.

Perhaps future economists will come to recognize these limitations and experiment with alternative ways of tracking inflation. They might turn to monetary aggregates, differences in interest rates that reflect inflation expectations, or survey-based measures of how people perceive price changes. None of these is perfect, but each offers a glimpse into changes in the value of money that a single representative basket of goods can’t fully capture.

Ideally, of course, we would have a stable unit of account, a monetary yardstick whose value doesn’t drift dramatically from year to year. In that world, nominal values would be enough. Real and nominal would coincide, and we could stop pretending that one person’s consumption choices speak for everyone.

Until then, every time we correct for inflation we might pause to ask, “Whose inflation are we measuring?”

If you ask economists what the optimal tax regime looks like, you’ll get a menu of reasonable answers. A flat income tax with a broad base and few carve-outs scores well on simplicity and growth. Estonia’s flat personal income tax (22% in 2025, set to rise to 24% in 2026) is a clean, modern example. Others favor consumption taxes (e.g., VATs or progressive consumption taxes) on the grounds that taxing spending rather than saving distorts growth less.

Those are good ideas. But if we’re talking about the ideal system, I would go further.

The optimal tax system is one in which there are no taxes at all. Government operations are financed from the returns on assets it owns, not from compulsory levies on income or consumption.

This isn’t as utopian as it sounds. In fact, you can already see the outlines of how the U.S. could get there.

Public Equity Stakes and Revenue Rights

Over the past year, Washington has quietly started behaving a bit more like an investor—sometimes clumsily, but directionally important.

• Semiconductors: The federal government took roughly a 10% stake in Intel as part of a shift from counterproductive grants toward equity stakes.

• Rare earths: The Department of Defense is moving to own about 15% of MP Materials, a key U.S. rare-earth miner, making the Pentagon a top shareholder.

• Lithium: Reports indicate that stakes have been taken, or discussions are underway, for the government to take 5 to 10% positions in major lithium ventures tied to U.S. supply chains.

• AI chips and export access: A novel deal allows Nvidia and AMD to sell AI chips to China in exchange for the U.S. government taking a 15% cut of related revenue (a revenue right rather than an equity stake).

Any single investment may or may not pay off. That’s not the point. The point is that these are unambiguous steps toward a government that thinks in terms of wealth creation. That’s the mindset change we need to start down the long road away from taxation dependence.

We Have the Assets

The U.S. federal government already sits on trillions of dollars in financial and non-financial assets. The Treasury’s annual Financial Report of the United States Government shows multi-trillion-dollar balance-sheet assets, including financial assets, loans, property, plants and equipment. Much of this portfolio is under-managed relative to its potential.

Beyond that formal balance sheet, the federal government possesses:

• Land resources. Washington owns about 640 million acres (about 28% of U.S. land).

• Mineral and energy resources. Federal lands and waters generated about $16.45 billion in FY2024 revenue alone.

• Spectrum. Periodic FCC auctions generate large one-off windfalls.

Why isn’t the federal government already a profit center? Because our institutions don’t prioritize financial returns. A sovereign wealth fund of the type the Trump administration might be creating changes that. Assets could be managed for return, subject to explicit constraints related to risk, national security, and other concerns.

“But You Need a Budget Surplus to Build an SWF.” (No, You Don’t.)

A common objection to a U.S. sovereign wealth fund is that countries need budget surpluses to seed a fund. That’s too restrictive. If an opportunity has a positive net present value, turning it down simply because the consolidated public sector is a net borrower leaves value on the table. The right test is a project-level assessment at the margin, not whether the overall government balance is negative.

Put differently, we shouldn’t reject positive-NPV investments just because the sovereign balance sheet is a mess. In fact, the only way out of the hole we are in is to start taking small, positive steps in the right direction. Portfolio building of the sort Trump is doing is how we dig ourselves out.

Some principles that could be followed in this endeavor include:

• Use tools like minority equity stakes, revenue rights, and royalty streams to better align policy with long-term value creation.

• Evaluate each potential investment based on risk-adjusted return and alignment with U.S. interests.

• Move to wall off the fund by giving it statutory or constitutional independence, a clear investment mandate, and GAAP-style reporting requirements so doesn’t turn into a political slush fund.

Critics warn that public stakes politicize markets. That risk is a concern, but experience shows it is overblown. The solution isn’t to keep running the government in a way that destroys value year after year. Rather, it is to adopt the safeguards already used by the world’s best funds, like Norway’s GPFG.

Safeguards could include oversight mechanisms such as having an independent board, holding passive or minority stakes in most equities, using competitive auctions for revenue rights in areas like spectrum and mineral leases, elevating financial fiduciary duty above any social or political goals, and requiring strict disclosure and regular audits.

Back to Taxes

History offers us some perspective. In much of medieval Europe, taxation was considered a last resort. Rulers were expected to live off their own estates and rents, and public opinion often viewed new taxes as illicit or parasitic. That norm eroded with the rise of the modern fiscal state, but the change need not be permanent. If anything, today’s federal portfolio is larger and more monetizable than any medieval estate.

If recent U.S. deals signal a shift toward financing government from assets, that’s progress, not because each investment is perfect and is guaranteed to reap high returns, but because the direction is right. It signals a shift in mindset away from taxing productive activity and toward earning returns on a professionally managed public portfolio.

The optimal tax system is no taxes. I don’t expect to get there in my lifetime. But starting to judge policy outcomes by whether they create net wealth pushes us firmly along the right path.

One of the core insights of economic theory, which might come as a surprise to some, is that free markets tend to produce suboptimal outcomes in terms of economic efficiency. This doesn’t mean markets fail in every respect. In fact, markets are quite good at coordinating decentralized activity and generating prosperity. But from a theoretical standpoint, even perfectly competitive markets fall short of maximizing the total economic surplus society could produce.

Here I use the term efficiency to describe this ideal of surplus maximization. Moreover, efficiency itself has two components: static efficiency and dynamic efficiency.

Static vs. Dynamic Efficiency

Static efficiency concerns the allocation of resources within a given time period. It relates to whether we are using what we have today in the way that best satisfies current preferences, as measured by people’s willingness to pay for goods and services. If all products are being produced and consumed according to what people want most, given their budgets, tastes and current technologies, we achieve static efficiency.

Dynamic efficiency, on the other hand, concerns how resources are allocated across time. It’s about growth, and whether we are saving and investing enough today so that the economy’s productive capacity expands in the future. Dynamic efficiency is about maximizing the size of the pie over time, while static efficiency is about dividing the pie optimally at each moment.

If both forms of efficiency could be achieved simultaneously, society would be as wealthy as possible across and within every time period.

Why Economists Focus on the Easier Half

In practice, most economists focus on static efficiency. It’s practical, predictive and lends itself to elegant models. Dynamic efficiency, by contrast, seems “too hard.” It involves difficult ethical and political tradeoffs related to how much the current generation should sacrifice for future generations.

Behind this underinvestment problem lies a deeper issue related to human nature. Individuals possess positive rates of time preference, which is our universal tendency toward impatience. Because humans generally prefer present consumption to future consumption, we discount the future heavily. The result of our nature is a market equilibrium with less saving and lower long-term growth than would be ideal.

Even if society could achieve the optimal rate of saving and investment to reach an optimal path of economic growth, that equilibrium would be unstable. Each generation faces a temptation to consume a little more for itself, leaving less for those who come after. Each can make itself better off in the short run at the expense of the long run.

So, in this respect, we live in a kind of ongoing market failure. The decentralized decisions of individuals, guided by market signals, fails to produce the level of investment that would maximize the wealth of society over time.

Three Responses to Market Failure

Given this diagnosis, what should be done about it, if anything? Here are some options.

1. Reject Efficiency as a Normative Ideal.
One option is to say, “who cares?” Efficiency, after all, is not the only thing that matters. People might value justice, equality, or beauty more than maximizing consumer and producer surplus. In that case, economics functions merely as a descriptive science, helping us understand how real-world markets work. It shouldn’t be used to make claims about what we ought to do.

2. Accept Suboptimality and Move On.
A second option is to acknowledge the inefficiency but decide it’s not worth fixing. The laissez-faire position, roughly speaking, holds that while markets may not be perfect, government interventions to correct them often make things worse. The cure, in this view, is likely to be more harmful than the disease.

3. Try to Fix It.
The third option is the most ambitious. Attempt to do something about the dynamic inefficiency in order to pass on a better world to our descendants.

If We Choose to Act

Suppose, for argument’s sake, we decide to intervene. What might that look like?

One possibility is to increase investment on behalf of future generations. This could take the form of a public or private intergenerational wealth fund, i.e., a pool of resources set aside to accumulate capital over time. Such a fund need not be run by the federal government. It could be state-level, local, or private, or with multiple public and private funds all competing to manage society’s savings most effectively.

But perhaps the solution doesn’t have to be governmental. Maybe it starts with individuals. People could choose to save a bit more, not just for themselves in their retirement or their children, but for society. Financial innovators could create new products to encourage this, like special long-term savings accounts or voluntary intergenerational funds designed never to be fully liquidated for consumption within a person’s lifetime.

AI also opens new frontiers. Imagine autonomous agents that own resources and continuously reinvest them, building wealth not for immediate use but for the indefinite future. Such mechanisms could gradually increase the investment rate and lower the real rate of return until it reaches its “optimal” level.

That optimal rate wouldn’t be zero, since the real interest rate reflects both time preference and risk. But if the time preference component could be reduced or eliminated entirely (through cultural influences, technology, or voluntary savings mechanisms) we could move closer to the dynamic efficiency ideal without coercion.

A Word of Caution

Of course, more investment doesn’t automatically mean lower interest rates. In cases of increasing returns, greater investment and output can raise productivity and returns at the same time. In such a world, there might be no natural limit to optimal investment beyond the bare minimum needed for people to sustain themselves.

These are speculative ideas, but they raise profound questions about what kind of future we want to build and what role markets can play in shaping it.

Closing Thoughts

I don’t claim to have all the answers. My aim here is simply to sketch the terrain of possibilities. Human nature being what it is, free markets, for all their virtues, systematically underinvest in the future. Whether we choose to accept that fact, ignore it, or try to correct it—privately or collectively—will determine how prosperous future generations become.

A better world is possible. The first step is recognizing that our present one, even at its best, is not what it could be.