Originally posted: ChrisB’s Blog

Productivity is a threshold. Positivity is a floor.

Chris Bland and Warren Mosler [1]

The economy of 2050

Assume the current trajectory roughly holds. Not the wilder claims about machine consciousness or the end of history, just the boring extrapolation: large language models keep getting cheaper and more capable, robotics improves steadily, if less dramatically, and both keep diffusing into ordinary work at the pace such things usually diffuse. Twenty-five years of that. What does the economy look like at the end of it?

A law firm in 2050 has a fraction of the juniors it had in 2020. The first-pass research, the document review, the drafting of things that have been drafted ten thousand times before, all of it is done in seconds by a system that costs less per month than one of those juniors cost per day. The partners are still there, because clients want a person to be accountable and a person to shout at. The pipeline that produces partners is not.

The same pattern runs across the professions. Translation is machine work, and most of radiology, and copywriting, entry-level analysis, first-line support, the junior end of software, the research assistant, the paralegal, the sub-editor. Not the senior people. The rung beneath them, which is where everyone used to start.

The physical economy is further behind and patchier. Warehouses are almost entirely machines, and so is long-haul freight, and the repetitive end of construction, and agriculture wherever the terrain allows. The trades are the holdouts, because a robot that can rewire the first floor of a Victorian terrace is a far harder machine to build than one that packs a box, and the volume never justifies it. Plumbers are doing rather well.

What survives is, by definition, the more important work: in this case, the work where a person is the point rather than the means. Care for the very old and the very young. Teaching, at least the part that involves noticing a child has stopped trying. Nursing. Hospitality, live performance, hairdressing, the long tail of physical tasks in awkward places. Anything where the customer is buying the presence of another human being, and knows it, gains in importance.

Underneath all of it sits something the forecasts tend to miss. The machines run on electricity, electricity runs on infrastructure, and infrastructure is made of things that take years to build. The binding constraint on new physical capacity in most rich countries is not capital and not labour but grid connection queues and a global shortage of high-voltage transformers, a component nobody thought to worry about because nobody had needed one in a hurry since the 1970s. The constraint has moved from people to power, and from power to the industrial base that makes the things that carry power.

One thing about that world remains genuinely undecided, and it will matter later. If frontier capability keeps arriving only through enormous capital and energy, then compute is a strategic asset, controlled by the handful of states that can build it and rationed to everyone else. If instead it arrives through cheaper architectures and published weights, capability diffuses, and anyone with a modest cluster and a power supply can have it. Call these the two branches. Almost nothing else in this paper depends on which one wins, and the last part of it depends on nothing else.

So the economy of 2050 is a continuation of the productivity gains of the past. Manufactured goods are absurdly cheap. Information is close to free. Housing, which sits on land that cannot be manufactured, is not as cheap. Care is not cheap: human attention stubbornly refuses to get cheaper, and is increasingly the only thing anybody spends money on.

And if a great many people cannot find paid work in that economy, with human attention so dear, it is evidence of a policy choice, even in an economy producing more than it has ever produced.

The case against the MMT approach

An economy of that shape is unsettling, but it is not obviously a crisis. Machines have been displacing workers since the spinning jenny, and each time the economy has eventually found more valuable things for people to do. There is no obvious reason to assume this time is the exception.

The reason to worry is narrower than mass unemployment, and it should concern MMT in particular. What the automated economy threatens is the set of instruments we use to tell whether an economy is in trouble at all. Four problems follow.

What comes next is the case against MMT put as forcefully as its critics could put it, and none of the four problems is trivial. MMT’s answers come later, in “The four problems, answered”, and they are not the obvious ones.

The measurement problem. Imagine a statistical release in 2050 reporting twelve per cent unemployment and ninety-nine per cent capacity utilisation. For two centuries those numbers moved together, and economists built an apparatus on the assumption that they always would. Idle workers meant idle capacity. MMT leaned on this harder than most: unemployment is by definition people looking for paid work, so offer them the currency and they will take it, which is direct evidence that real resources sit idle. But that inference appears to hold only while labour is the marginal input, the thing that lets you produce more when you add it. If the constraint is transformers on a three-year lead time and fabrication capacity that cannot be conjured, the people are real and they tell you nothing about whether the next project can be built.

The anchor problem. The job guarantee is often mistaken for welfare policy. In the theory it does two things. It eases the transition from unemployment to private sector employment, and it acts as a price anchor: a fixed wage that no private employer can undercut, providing a wage floor for an input that enters nearly every cost structure, with a quantity that expands and contracts as the private economy sheds and hires. The objection is that once labour drops out of the cost structure, the floor stops propagating. You could hold the guaranteed wage constant while every price in the economy moves.

The tax problem. MMT is chartalist. The currency has value because the state imposes a liability denominated in it and enforces it. The liabilities modern states actually impose are built on wages and transactions. Follow the automated economy through and output is produced by capital, owned through structures whose most valuable asset is intellectual property, which is the most mobile asset ever devised. The circuit then depends on taxing the most mobile actors in the world economy, by states competing with each other to host them.

The distribution problem. Output can be abundant while income is concentrated, because the returns accrue to whoever owns the capital and the intellectual property. If wage income shrinks as a share of household income, the mechanism that spread the gains of every previous productivity wave is gone.

All of this may of course be premature. We have had capable large language models for years and measured productivity growth remains unremarkable. Robert Solow, the Nobel laureate who worked out how much of economic growth comes from technology rather than from adding capital and labour, observed in 1987 that you could see the computer age everywhere except in the productivity statistics. That gap ran for a decade and a half before the computers finally showed up in the numbers, and it may be running again. Demographic ageing pushes the other way too, shrinking working-age populations across the OECD and China while the demand that explodes is care work, the hardest thing to automate. If several of those hold, this is a normal productivity wave with normal reabsorption and the question answers itself.

That is the case at its strongest. All four problems dissolve on contact with the framework properly understood, which is not the same as saying nothing survives. What survives is a fifth problem, and it is not a monetary one.

Why neoliberalism has no answer

Neoliberalism’s claim, as mainstream economists present it, is not merely that markets allocate efficiently. It is that market wages reflect marginal contribution, and that this is therefore just. A person’s income is the measure of what they add. On that account every remedy for displacement takes the same form: get the worker back over the productivity line, the point at which what they add is worth more than what they cost. Retraining, wage subsidies, competitiveness, human capital investment, labour market flexibility. All of it presupposes the line can be reached, and that a cheaper worker is a more employable one.

The deeper flaw is that this analysis is based on the dynamics of a fixed exchange rate policy, which is entirely inapplicable in today’s world of non-convertible currency and floating exchange rates. Under a fixed rate, a government must defend convertibility, and unemployment becomes part of the price of doing so. Under a floating rate that constraint is gone. But the position fails even on its own terms.

Firms do not actually work the way the doctrine assumes, which makes the position worse rather than better. A firm does not hire up to the point where the last worker’s revenue matches their wage. It works out how much it expects to sell, employs the people needed to produce that, and sets prices as a markup on its costs. Fixed capital dominates those costs, plants are built with slack in them, and the cost of one more unit is usually close to nothing. So employment is set by expected demand and by what the technology requires, not by a marginal calculation at the factory gate.

That removes the escape route the doctrine relies on. In the marginal story a wage cut eventually brings the worker back inside the line, and that is what licenses every remedy listed above. Under demand-determined employment there is no line to come back inside. A firm that needs forty people to make what it expects to sell hires forty people, and no wage is low enough to make it hire the forty-first. When the machine does the work, the requirement falls and the jobs do not come back at any price.

So the justification is left standing on its own, with nothing underneath it. The claim that wages measure contribution, applied to an economy where machines have taken over most of the contributing, becomes an argument for the destitution of the people it describes. And it has no reply, because the reply would have to be that people should be paid for reasons other than what they produce, which is the one thing it cannot say.

It would be too much to claim that automation kills capitalism. Private ownership of capital and production for profit survive automation comfortably and may thrive on it. What breaks is the distributive settlement that made the arrangement politically tolerable. For two centuries capitalism could say that it paid people because it needed them. Remove the need and the payment has no basis inside the system. The mechanism survives, the legitimating story does not, and the gap between them is the political crisis of the period.

Which is why the alternative its defenders reach for concedes the whole argument. A universal basic income pays people precisely because they are not needed. It is an injection with no fixed price and no elastic quantity, so it stabilises nothing. And in an economy where land is the residual scarce asset, a cash transfer against fixed land supply is absorbed into rent, which makes it a subsidy to landlords routed through the population. It is not a distributive answer. It is a way of avoiding the question of distribution altogether.

More fundamentally, however, UBI can put the state’s ability to provision itself at risk. The state imposes tax liabilities so that people need its currency, and the state is the sole supplier of it, so people work for the state to obtain what they need to pay their taxes. If the state simply hands that currency out as UBI, it removes the need to work for it, which markets express as hyperinflation.

Productivity and positivity

One sentence reorganises the whole argument. Warren Mosler wrote it in reply to a question about whether the framework permits an economy in which the marginal product of additional labour is at or near zero across most sectors.

Depends on how you measure the real marginal product of labor. It only needs to be positive to be worthwhile?

Near zero is not zero. That is the whole thing.

The private employer applies a threshold, whatever the textbook says the threshold is. It hires the people it needs to make what it expects to sell, and every advance in the technology lowers that number. The currency issuer applies a floor instead: if the work produces anything real and the public purpose is served, the work is worth doing. It does not have to beat the machine. It has to be better than nothing.

That is a far larger claim than its phrasing suggests. It says machines cannot render a person economically worthless, because worthlessness is an artefact of the threshold, and the threshold is a private-sector convention rather than a fact about the world. Between a threshold and a floor lies the entire difference, in 2050, between a functioning society and a redundant population.

It also does something a purely operational framework is not usually credited with. It supplies a moral position that is not bolted on. MMT does not need to argue that people have dignity beyond their productivity. The proposition falls straight out of the description of how a currency issuer works.

Ask Mosler where the work comes from and the answer is not a productivity story.

There’s lots that might not be possible, but lots more that is possible and is useful. How many research assistants would university professors hire if they had the funding, for example? Even something simple like holding hands at old age homes?

Neither of those clears a productivity threshold. Both clear a positivity floor.

The four problems, answered

With the floor in place, all four look different, and none of them looks like a hole in the theory.

One: the gauge was never measuring what it appeared to

The “measurement problem” was built on the idea that unemployment works as a gauge because adding a worker adds output. The better starting point is that full employment serves public purpose in itself, not least by reducing the negative externalities that unemployment creates. The correction concerns who the unemployed actually are, and where they came from:

It’s only a gauge of whether those the tax liabilities have caused to be unemployed have been hired by the state imposing those liabilities. That is, the unemployed have been shifted from private to public domain by state tax liabilities. Note that with that understanding, the non gov. sector remains fully employed. It’s the currency monopolist alone that causes unemployment, as defined as those looking for work that pays in that currency.

Unemployment on this account is not a symptom of a weak economy. It is an artefact of the state’s own arithmetic. A tax obliges people to obtain the currency, and for most of them the only way to obtain it is to sell their labour for it, so the liability is what puts them into the market looking for paid work in the first place. Having created that queue, the government then buys from part of it. The part it declines to buy from is what the statistics record.

Private firms sit outside the frame entirely. They have hired whoever they want at the wages on offer, and by their own reckoning they are finished. Nothing they do or fail to do produces the number.

So the paradox was never a paradox. Twelve per cent unemployment tells you the government has left a fiscal gap open: a failed policy response, readily addressed by appropriate fiscal adjustment. Ninety-nine per cent capacity utilisation tells you there are no spare transformers, and nothing about the more important question of what services people could be doing to benefit each other. Both readings are accurate, and they were never measuring the same thing. One number appeared to do both jobs for two centuries only because labour happened to be simultaneously the abundant input and the binding one. That coincidence is what ends. Not the gauge.

Behind it sat a word doing double duty. Where the “measurement problem” suggested unemployment would stop being evidence of slack, Mosler’s reply was:

Depending on how you define slack. It’s a casual term.

The ambiguity had run through everything. Slack was doing two jobs at once: idle resources the state could put to work, and headroom to increase output of a particular constrained thing. Unemployment measures the first reliably and is silent on the second. Every apparent paradox in the case against MMT was the same ambiguity in a different costume.

What the automated economy genuinely requires is a second instrument alongside the first. Idle compute, spare grid interconnection, industrial capacity utilisation by sector, materials inventories, lead times on physical capital. Not a replacement dashboard. An additional one, reading the physical constraint that unemployment was never measuring.

Two: the anchor was never in danger

The “anchor problem” held out a little longer. In an economy where machines are the first choice at nearly every stage, human labour gets used last if at all, so adding a worker adds nothing.

“You assume the gov. doesn’t need labor?”

The objection was about private firms and their production functions. But the buyer that matters here is not a private firm. It is the state, and the state does not hire at the margin of profitability. It hires to serve public purpose. Public employment is always an option, and there are always more public services that can add to quality of life than there are people to provide them. Government demand for labour is not a function of what a machine could do instead, which is exactly why a standing offer from an entity that is not optimising survives in a world where every optimiser has stopped hiring.

The anchor holds for the same reason. A price anchor works by fixing something the government pays for and letting the quantity float, and the general price level then takes its bearings from the prices the government sets. That mechanism does not require labour to be the marginal input anywhere in the private economy. It requires the state to keep paying the same wage to anyone who turns up, and to keep doing so when the private sector is booming as well as when it is not. The state imposes the tax liability, and the job guarantee wage then tells the economy what it must do to obtain the currency needed to pay it.

There is a second reason the anchor holds, and it is the more interesting one. Automation compresses labour’s cost share unevenly. Machine-produced goods get cheap. What stays expensive is housing, energy for physical processes, food, and the services where a human being is the product rather than an input to it: care, teaching, hospitality, the trades in the long tail. Those are precisely the categories that generate price volatility, and in most of them labour remains the dominant cost.

This is Baumol’s cost disease [2], the old observation that a string quartet needs the same four players it needed in 1800 while everything around it gets cheaper, so the relative price of the quartet climbs. Usually cited as a problem. Here it runs in the direction that helps. The job guarantee may end up covering a smaller share of employment while anchoring a larger share of the prices that actually move. Fewer people, more leverage.

So the floor does not stop propagating when labour leaves the cost of manufactured goods. It continues through the service sector. And even if the guaranteed wage held constant while other prices moved, the population would be fully employed and output fully consumed as markets discover relative values. It is the same observation as the research assistants and the old age homes, arrived at from the other end.

Three: the tax base moves to what cannot move

The “tax problem” assumed that tax must follow income wherever it goes. The answer is blunt. Mosler notes that tax exists to

create sellers of goods and services for the further purpose of provisioning government. The rest of the institutional structure serves the additional desired public purpose.

One purpose, with everything else following from it. In a framework this precise about operations, the distinction is the whole game.

It also disposes of the problem. The tax liabilities a state imposes are exogenous decrees of its legislature, not facts of nature. That today’s are built mainly on wages and transactions is a choice. If the tax exists to create sellers so government can provision itself, the question is not how to chase mobile income around the world. It is what the state can impose an enforceable liability upon. The answer is whatever cannot leave: land, property, energy at the point of generation or import, and the physical right to operate on national territory. That points towards asset taxes, such as taxes on real estate, rather than transaction taxes such as income and sales taxes. Nothing about the automated economy makes that harder. If anything it makes the immobile base more valuable relative to everything else.

Four: rent, which MMT had already thought about

Which brings us to the “distribution problem”. The first point is that distribution is a function of institutional structure, which can be altered at will. When wage income shrinks, the dynamics change, but the means of spreading the gains does not disappear. It has to be redesigned.

If the gains from progress are absorbed by whoever owns the land, then a land value tax (on the grounds Henry George set out in 1879), would seem the way to go; functional finance suggests that the tax need not raise revenue and can be set at whatever rate best suppresses rent.

Mosler’s reply went back to what the land value tax was originally designed to do:

It was meant to promote land development, which was last year’s battle. Today we need to control property development, hence a property tax with ‘exemptions’ for desired development is the actual solution.

And then the analogy that explains why. Bank regulation tells banks what they cannot do, and regulators then spend their careers stamping out the new bad things that appear and adding them to the list. Tell banks instead what they can do, and let them petition for additions. A property tax with exemptions is that structure applied to land: development discouraged by default, permitted where wanted. A land value tax is the other structure, a general incentive to develop followed by an endless effort to suppress the development you did not want.

The historic objection to taxing property values was never really theoretical. It was administrative. Separating land value from improvement value across millions of parcels was slow, expensive and endlessly contestable, and the assessment burden killed the policy politically more often than any argument against it did. Mass appraisal is exactly the sort of problem that cheap abundant intelligence solves. The technology that prompted this whole exercise makes the instrument newly practical. So distribution is not the orphan it appears. It has an instrument, aimed at the asset that a machine economy makes scarcer rather than cheaper, and for the first time in a century that instrument is administratively feasible.

What the framework is for

The framework is built on operational description. MMT is not a predictive model. It is an account of what happens in the payments system when a government spends and taxes, and accounts of what happens do not become false because the technology changes. Hand it any future and the logic holds.

There is a cost to that, worth naming without turning it into an accusation. A framework that absorbs every scenario without updating is difficult to test, and if no imaginable future requires its terms to change, it is correspondingly hard to say what would count as evidence against it. That is not fatal, because descriptive claims are the right claims for it to be making. It is simply what you notice after trying hard to break something and watching it decline to break.

There is a moral architecture underneath the operations, and it surfaces in the observation that unemployment remains a policy failure:

Always. It’s a government imposed crime against humanity.

The alternative is not about growth or competitiveness or productivity.

We wake up every day with a massive real labour shortage but unfortunately there’s a politically engineered shortage of funds to realise the potential.

That is the whole position. There is more that needs doing than there are people to do it, and always has been. The constraint is not technology, or resources, or in most cases physics. It is an accounting convention we impose on ourselves and then mistake for a law of nature.

Machines make that convention more expensive every year. If they deliver what is claimed for them, potential output rises enormously, and the gap between what is possible and what actually happens becomes a purely political artefact. The automated economy does not threaten MMT. It removes the last respectable excuse for ignoring it.

The fifth problem

One problem is left. MMT recognises it, but cannot resolve it on its own.

Start with the base case, which is correct and not in dispute here. Exports are a real cost and imports a real benefit, because you produce something and somebody else consumes it. MMT’s external answer follows from it: with a floating exchange rate there is no financial constraint on a currency issuer, only a real one. A country can always buy what it needs in its own currency at some price. If it has little to sell, the terms are punishing and the country is poorer in real terms. That is an unpleasant answer and a complete one, and it has covered food and fuel for as long as anyone has imported them.

Now add compute, and the energy to run it, to the list of things an economy cannot function without.

MMT’s boldest claims have always carried a quiet condition. A currency-issuing government faces no financial constraint in its own currency, provided it does not need to buy necessities in someone else’s. That condition has never been trivial. It is why the framework speaks confidently about the United States, Japan and Britain, more carefully about Brazil and India, and hardly at all about countries that import their staples.

And here the two branches matter, which is why they were worth flagging at the outset. On the diffusion branch, where capability is cheap and the weights are published, compute is a priced input like any other and MMT’s external answer works normally. On the enclosure branch it is not priced at all. Compute is dual use and militarily decisive, and states do not sell the things that determine whether they win wars. The plausible regime there is not a market but something shaped like nuclear non-proliferation: export controls, licensed access, capability tiers written into alliance structures.

That regime exposes the one thing a floating currency cannot do. It handles scarcity rationed by price. It has no answer to scarcity rationed by permission. You cannot devalue your way into buying something that is not for sale to you at any exchange rate, and that, at least, looks new.

Even this is less new than it looks. MMT begins with the currency as a public monopoly, and with what it means for everyone else to face a single supplier. A state that rations compute by permission is a monopolist too, and the encounter between the two is what Mosler has called a case of duelling monopolies, like the contest between OPEC and the currency issuers that buy its oil. The framework recognises the shape of the problem. What it cannot settle by itself is the outcome, which turns on power rather than price.

Follow the enclosure branch and the world sorts itself into tiers that differ in kind rather than degree.

At the top, a very short list of states holding fabrication, frontier capability and the power to run them, with no obvious reason for the list to lengthen.

Below them, the energy-rich and compute-poor, whose position is stronger than it first appears. Compute has to be sited where the power is, and that is a bargaining position rather than a dependency. It is not oil, which could be shipped away leaving the country holding a royalty and nothing else. Data centres have to be built somewhere and connected to a grid somewhere.

Then the renting middle, monetarily sovereign and technologically tenant. Able to issue its own currency, unable to use it to buy the thing its economy increasingly runs on except from a supplier who may attach conditions.

And at the bottom, the countries with neither, where MMT’s domestic argument stays technically true and practically inert, in the way it is technically true of a country that must import its wheat.

So the open questions are these. Whether a country can obtain the compute and energy its economy runs on without needing somebody else’s currency to do it. Which states end up holding the physical layer, and what they charge for access, in money or in alignment. Whether the international arrangements that might prevent that concentration can be built by an order coming apart for the same reasons, since the concentration of capability is itself one of the forces dissolving multilateralism, handing a few states leverage they have no incentive to pool.

These are questions about the conditional clause MMT carries, so they are within its scope. But they resolve into international political economy rather than monetary theory, because the binding mechanism is alliance structure and export control rather than price. Trade theory is built for a world in which the thing you need is for sale. It has nothing to say about one in which it is withheld on purpose.

MMT does have a partial answer, and it is the right one. If you cannot buy the capability, command the resources to build it. That is what a currency issuer is for. It is the argument for public compute provisioned as a utility at an administered price, universally, on exactly the architecture of the job guarantee: a price the state sets, a quantity that floats, and access that is not rationed by ability to pay. A state that provisions compute publicly has a floor under its capability that a state renting from a foreign supplier does not. And if the longer-term outlook in this paper holds, with labour growing dearer relative to everything else, the quality and quantity of a country’s domestic labour force will translate into ever more favourable real terms of trade.

The limit is that fiscal capacity does not conjure fabrication plants inside three years, and the countries best placed to build are the ones already holding the energy and the industrial base. That is not a monetary problem. It is a question about who holds the physical layer of the world economy and what they choose to do with it, and nobody has a good framework for that.

Conclusion: Power, not money

This paper suggests a solution to the four problems above and stresses that unemployment measures the fiscal gap and always did. The job guarantee needs labour to be useful, not to be marginal. The tax creates sellers so the state can provision itself. Slack is a casual term.

What the automated economy actually does is sharpen the choice the framework has always described. When machines can produce nearly anything, the distance between what a society could do and what it does becomes entirely a matter of political decision. And a doctrine that says people are worth what the market will pay them becomes, in an economy where the market will pay most of them nothing, an argument for discarding most of the population.

Set against that is a framework that asks a different question about a person’s work. Not whether it clears the productivity line, but whether it is worth doing at all. The research assistant a professor cannot afford. The hand held in an old age home. Neither will ever beat a machine on cost and neither has to, because the currency issuer is not shopping for the cheapest supplier. It is deciding what is worth paying for. A framework built on that question, which treats unemployment as a crime committed by governments against their own people, does not look like a heterodox curiosity. It looks like the only account of the economy that survives contact with abundance.

What remains unresolved is not internal to it. It is whether the states that will need to command those resources are permitted to obtain the inputs, by other states who have worked out that capability is the one export worth withholding. That is a question about power rather than about money, and monetary theory, having answered the questions it claims, hands it on.

Notes

[1] The idea for this paper came from Chris working with Claude (Opus 5.5 High), which also helped with some editing. Warren added his thoughts to Chris’s work (his direct quoted responses are inset and italicised). The resulting paper is a deeply thought-provoking view of the near future. Phil Armstrong

[2] See Baumol, W. J. and Bowen, W. G. (1965). “On the Performing Arts: The Anatomy of Their Economic Problems”. The American Economic Review, 55 (1/2): 495–502. JSTOR 1816292.