Recent studyBlind TTS Elo is live. Compare two anonymous voice samples, vote after listening, and help separate real preference signal from noise.Vote in the study ->
Home/Explainers/The Deflationary Doom Loop
Economics / AI · A scenario, not a forecast

The Deflationary Doom Loop

When chatbots become agents, the layoffs accelerate — and a frightened workforce starts to hoard. Here is the macroeconomic trap that could follow, and the one physical fact the utopian story keeps forgetting.

An economy spiralling into a vortex of falling banknotes and office chairs
The spiral: value swirling inward as spending contracts. Illustration generated with Google nano-banana-2.

Epistemic status: a serious thought experiment, not a prediction. The argument chains together mechanisms economists already understand. Whether they actually combine this way is the open question — the counter-case gets its own section at the end.

Adapted from a thread by @shdu11546816 on X ↗.

§ 01 · The engine

The Paradox of Thrift

Saving is virtuous for a household and dangerous for an economy. This is one of the oldest counter-intuitive results in macroeconomics: when everyone tries to save more at the same moment, total demand falls. Lower demand means lower revenue; lower revenue means layoffs and wage cuts; lower incomes frighten people into saving even harder. One person’s spending is another person’s wage, all the way down.

The loop is not a metaphor. It is an accounting identity wearing a human face.

Diagram · The self-reinforcing cycle
THE DOOM LOOPeach turn tightens the next1Fear of the futureA credible threat to your job appears.2Households hoard cashPrecautionary saving spikes.3Aggregate demand fallsLess spending across the economy.4Revenue dropsFirms cut payroll and wages.5Incomes fall…which deepens the fear.

Drag the slider below to raise the savings rate — to dial up the fear — and watch output spiral down to a new, smaller equilibrium. The model is deliberately simple: a Keynesian income-expenditure multiplier where the economy settles at output = autonomous spending ÷ savings rate.

Simulator · Everyone tries to save more at once
10% · calm25%45% · panic
New equilibrium
143
64% below full
Multiplier (1/s)
3.6×
shrinks as fear rises
Total saving, $
40
unchanged — the paradox

The cruel twist: when everyone hoards at once, income falls until the dollars actually saved are the same as before — the economy just runs smaller and poorer. Individual prudence becomes collective self-harm.

For ninety years this stayed mostly theoretical, because mass simultaneous panic was rare and brief. What if a single technology could trigger it across the entire white-collar economy at once?

§ 02 · The trigger

Chatbots were a demo. Agents are a workforce.

A chatbot answers. An agent acts — it plans, executes, calls tools, and closes the loop without a human in the seat. That distinction is the whole ballgame. A tool that assists a worker raises their output; a tool that replaces the worker removes their paycheck. The first is a productivity story. The second is a demand story.

Knowledge workers at their desks dissolving into AI-agent forms
The quiet handover: cognitive labor migrating from people to agents. Illustration generated with Google nano-banana-2.

The exposed cohort is not the usual one. Previous waves of automation hit factory floors and loading docks. This wave reaches first for the people who were told they were safe: software engineers, lawyers, analysts, copywriters, marketers, middle managers — high-income knowledge workers whose entire job is producing text, code, and decisions. They are also, not coincidentally, the people with the most discretionary spending to withdraw.

The first dominoes to fall are the ones holding up the most demand.

§ 03 · The relief valve fails

The Physical-Labor Glut

“Just learn a trade” is the standard reassurance — robots can’t fix a leaking pipe or turn an elderly patient. True. But it mistakes a personal escape route for a collective one. If a handful of displaced analysts retrain as electricians, they do fine. If millions do it at once, they are not escaping the glut — they are the glut.

A vast queue of suited professionals holding plungers and toolboxes outside one small job site
Supply, meet inelastic demand. Illustration generated with Google nano-banana-2.

Demand for plumbing, wiring, construction and elder care is real but inelastic: there are only so many homes to rewire and patients to lift, and that number does not jump just because the labor supply tripled. Pour displaced cognitive workers into a fixed pool of physical jobs and the supply curve marches right while demand sits still. Wages go where they always go in a glut: down.

Simulator · The market for physical labor
Quantity of physical labor →Wage →subsistence floorDemandSupply
Clearing wage
38
index, 100 = pre-AI median trade wage
vs subsistence
23 pts
cushion shrinking

Demand for plumbers, electricians and carers is real but inelastic — there are only so many pipes to fix. Pour millions of refugees from cognitive work into that fixed demand and the supply curve marches right. The wage slides toward the only floor left: enough to eat.

§ 04 · The bottom

The Wage-Floor Collapse

Push that slider far enough and the clearing wage settles onto a floor — roughly the cost of calories and a crowded room. Not because employers are cruel, but because when ten people will do a job for the price of survival, survival becomes the price. Meanwhile the same fear that follows the layoffs makes everyone who still has income hoard it. Discretionary spending — restaurants, travel, the second car, the renovation — is the first thing to go.

A crowded single room with many people on mattresses and a single bowl of food
Subsistence: enough to eat, a place to sleep, nothing left over. Illustration generated with Google nano-banana-2.

And there is the loop, completed. Layoffs cut incomes. Fear cuts spending further. Lower spending cuts revenue, which cuts more jobs and more wages. Each turn tightens the next. Except this time the trigger is not a financial shock that passes — it is a permanent change in what human labor is worth.

§ 05 · The utopian counter-story

“But everything becomes cheap and abundant.”

This is the optimist’s rebuttal, and it is half right. When intelligence costs almost nothing, anything bottlenecked on intelligence gets radically cheaper — software, design, analysis, tutoring, diagnosis, entertainment. Deflation in those goods is real and wonderful.

The flaw is in the word everything. Look at where a working household’s money actually goes.

Chart · A typical household budget, by what AI can actually cheapen
33
17
13
8
12
17
71%
Land, energy, calories, human care. Intelligence is not the bottleneck here.
17%
Manufactured goods, software, media — the slice AI deflation truly compresses.

Shares approximate the U.S. Bureau of Labor Statistics Consumer Expenditure Survey.

Two-thirds of the budget is rent, transport, food and care — things made of land, energy, calories and human hands, not thought. Cheaper intelligence barely touches them. So the optimist’s deflation and the pessimist’s deflation are two different prices moving in opposite directions:

Chart · Two prices, opposite directions (log scale, 2024 = 100)
Cost of intelligence — collapsesCost of rent, energy & calories — does not

“Everything becomes cheap” is true — for everything that is made of thought. The line that matters for survival is the flat one. Your landlord does not accept tokens.

§ 06 · The hard limit

The Thermodynamic Wall

Intelligence is not the binding constraint on physical resources. You can think as hard as you like and the laws of physics do not move. This is the wall the abundance narrative runs into:

A glowing AI brain pressing against a stone wall carved with symbols of land, wheat and fuel
Cognition meets the conservation laws. Illustration generated with Google nano-banana-2.
Land

AI cannot print more acreage in desirable zip codes. Location is rivalrous by definition — the supply of 'near the good jobs and schools' is fixed.

Energy

No amount of cleverness repeals thermodynamics. A gallon of gasoline holds a fixed amount of energy; efficiency has hard physical ceilings.

Calories

Food is captured sunlight. Photosynthesis is ~1–2% efficient and arable land and fresh water are finite. You cannot reason a harvest into existence.

Software eats the world only where the world is made of software. Rent, fuel and food are made of atoms, and atoms remain stubbornly scarce no matter how smart the machine asking for them.

§ 07 · The split

A K-shaped world

Put it together and the economy does not rise or fall as one — it splits. The owners of the machines, the land, and the capital capture the gains from cheap intelligence. The people who used to rent out their minds for a living watch the price of that rental collapse while the price of survival holds.

Chart · Whose income rises, whose falls (2024 = 100)
Those who own the machinesThose who rented out their minds
A divided landscape: a small group in a sunlit estate above, a vast grey crowd below
The two-tier outcome: a small capital class above a large redundant one. Illustration generated with Google nano-banana-2.

At which point the essay’s most uncomfortable phrase has to be said out loud. The fear is that a large share of the population — the original claim puts it at two-thirds — becomes what the brutal old euphemism called “useless eaters”: people who consume but, in pure market terms, no longer produce anything scarce enough to be paid for.

The phrase is deliberately ugly — its origin is a Nazi term, unnütze Esser. It is used here precisely because the condition it names is morally intolerable. People are not “useless”; a system that prices them at zero is the thing that has failed. That moral revulsion is the point, not a footnote.

Scenario · Share of the workforce without market-clearing value

retains scarce, paid value  ·  economically redundant. Two-thirds become economically redundant.

§ 08 · The case against the doom

Where this argument might be wrong

A scenario you can’t argue against isn’t analysis, it’s prophecy. Here are the strongest reasons the loop might never close — and why each is contested.

Objection 1
The Luddite fallacy

Every prior automation panic was wrong. New jobs we can't yet imagine always appeared.

The catch

Every prior wave automated muscle and routine while leaving cognition to humans. This wave targets cognition itself. 'Something new will appear' has held for 200 years — but the thing that always appeared was a job for the human mind. That is exactly the input now in surplus.

Objection 2
Jevons & induced demand

Cheaper cognitive work means we'll simply do far more of it — more code, more medicine, more art — re-employing people at higher volume.

The catch

Jevons applies to the resource that got cheap (compute), not necessarily to the human. If the agent does the extra work too, demand for the input explodes while demand for the human stays flat.

Objection 3
Redistribution: UBI and ownership

If capital captures all the gains, tax it. A universal basic income or sovereign wealth fund pays everyone a citizen's dividend, demand is restored, the loop never starts.

The catch

Mechanically plausible — this may be the actual exit. But it is a political choice, not an economic certainty, and it must be enacted before the demand collapse, not after. The doom loop is fast; legislation is slow.

Objection 4
Energy and atoms get cheap too

Cheap intelligence accelerates fusion, robotics, vertical farming and construction — eventually collapsing the price of land, energy and calories as well.

The catch

The strongest optimist case. It hinges entirely on timing: physical-world R&D moves at the speed of permits, supply chains and thermodynamics. If atoms get cheap a decade after minds do, the gap in between is the doom loop.

§ 09 · The question

The race that decides everything

Strip away the drama and the whole scenario reduces to a race between two curves you have already seen on this page: how fast the cost of human labor falls versus how fast the cost of physical necessities follows it down. If atoms get cheap roughly when minds do, the optimists win and we inherit abundance. If there is a lag — and physics, land and politics all suggest a lag — then the gap between those two curves is where a third of the workforce lives, with collapsing wages and rent that won’t move.

The Paradox of Thrift turned a private virtue into a public trap once before, in the 1930s. The uncomfortable thesis of this page is that intelligence too cheap to meter could pull the same trap shut — not because the machines are cruel, but because demand is made of paychecks, and paychecks are exactly what is being automated away.

Abundance for things made of thought. Scarcity for things made of atoms. The danger is that most people spend their lives buying atoms.

Source

The original argument — “deflationary doom loop,” the physical-labor glut, and the “useless eaters” framing — comes from a thread by @shdu11546816 on X ↗. This page develops it with interactive models and the counter-case.

← All explainersNext: Grossman-Stiglitz Paradox →