When every firm automates,
everyone loses customers.
Falk & Tsoukalas show that AI displacement is a coordination failure. Rational firms, racing each other to automate, destroy the very consumer demand they depend on. Wage cuts, UBI, upskilling, equity, none of it fixes the trap. Only one policy does.
What this paper actually says.
Skip the equations. Here's the argument in language anyone can hold.
Imagine you run a company. You can replace some of your workers with AI tomorrow and save money. The decision looks obvious. But those workers were also buying things, your competitors' things, and sometimes yours.
Now imagine every company in the economy is making the same calculation on the same day. Together they wipe out a chunk of consumer demand. Each firm acted rationally. Each firm is now worse off. So are the workers. So, eventually, are the owners.
That is the AI Layoff Trap: a textbook market failure where private incentives and collective welfare point in opposite directions, and the usual fixes don't reach the root cause.
Five claims that hold up the whole argument.
Tap any card to expand the longer version. These are the load-bearing ideas.
How they build the trap, step by step.
A competitive task-based general-equilibrium model, walked through as a story.
- 1Baseline
Workers earn wages. Wages become consumer demand. Firms sell to those consumers.
- 2AI arrives
AI can do a growing share of tasks cheaper than humans, for any single firm.
- 3Firm decides
A rational firm sees automation lowers its own cost. It automates.
- 4Externality
Each firm ignores that displacing workers also shrinks the pool of buyers.
- 5Arms race
Competitors must match. Everyone automates faster than the economy can reabsorb.
- 6The trap
Aggregate demand collapses. Both workers and firm owners are worse off than before.
The lever every firm is pulling.
Drag the slider. Notice how profits rise, then fall, even as wages keep collapsing.
Slide right and watch profits peak, then collapse. That peak is what no individual firm can hold onto, because each one's best move is to keep pushing right.
Illustrative, not from the paper's numerical calibration, the shape of the divergence is the point.
If the paper is right, here's what to expect.
Labor's share of national income continues a multi-decade decline, now accelerated by capability gains in AI.
Firms that own the models and the compute capture more of the surplus; downstream firms become price-takers.
Governments backfill lost demand with transfers, but without addressing the externality the bill keeps growing.
Conversation moves from 'is AI taking jobs?' to 'who pays for the demand AI removes?'
AI-cited layoffs, mapped onto the paper's logic.
Toggle a lane to see which companies the paper's framework touches in each direction.
Tap a lane to highlight the companies whose layoffs map onto it.
Said its AI agent does the work of 700 customer service agents and cut headcount via attrition.
Klarna is the canonical demand externality story. The savings stay on its P&L, but the 700 lost paychecks were demand for everyone else's products. Klarna later admitted it was rehiring humans for quality, a hint that the private gain was thinner than the announcement implied.
CEO publicly tied HR and back office hiring freezes to expected AI substitution.
IBM is doing the arms race move in slow motion. It is not firing in bulk, it is letting attrition do the work while pointing AskHR and watsonx at every refill request. The paper predicts exactly this: the path of least resistance for a competitive firm is a quiet, permanent ratchet down in headcount.
Repeated 2025 waves hit cloud sales, customer success and legacy hardware teams while Oracle poured capex into AI data centres for the Stargate buildout.
Oracle is a textbook case of the paper's capital share shift. Cash flow is being redirected from human salaries into GPUs, real estate and power contracts. Workers who built the install base are cut while the same firm signs multi billion dollar AI infra deals, concentrating surplus in compute owners exactly as the model predicts.
Two back to back 'business optimization' programs in Waterloo and Bengaluru, framed around an 'AI first' operating model and Aviator product push.
OpenText shows the indirect channel hitting the enterprise software middle. Cuts skew toward support, services and middle management, the layer most exposed to copilots. The savings are explicitly being recycled into AI R&D, so the firm internalises the upside while the demand loss lands on the broader Canadian and Indian tech labour markets. This is the externality the paper says no single firm has an incentive to price in.
Content and translation contractors replaced by generative models; doubled down on 'AI first.'
The contractor channel is where the trap bites first. Contractors have no severance, no equity, no political weight, so they absorb the displacement silently. Their lost income is pure leakage from consumer demand, with none of the institutional friction that slows full time layoffs.
Repeated rounds framed around 'AI efficiency.' Heavy capex on AI infra, leaner human teams.
Alphabet is both the cause and the beneficiary in the paper's framework. It sells the automation tool to every other firm in the model, and it uses it on itself. The result is a double concentration: surplus flows to Google as a vendor, and within Google it flows to capital owners rather than the laid off engineers and recruiters.
Restructured around AI tooling; recruiting and middle management hit hardest.
Meta proves the arms race claim in reverse. The market rewarded the cuts with a stock rerating, which then forced every peer CEO to demonstrate similar 'efficiency.' Once one firm is celebrated for cutting, the Nash equilibrium shifts and the rest must follow or be punished by investors.
Agentforce deployment cited as reason fewer human support agents are needed.
Salesforce is selling the agent and using it. Marc Benioff openly said Agentforce is why hiring freezes hold. The paper warns this becomes self reinforcing: every Salesforce customer who buys Agentforce then cuts its own support team, removing demand from the very SaaS market Salesforce sells into.
From a 53-page proof to a CEO's all-hands slide.
How the paper's abstract claims connect to the specific decisions being announced right now.
Each CEO frames cuts as 'efficiency' or 'AI-first.' The paper says: in a competitive market they have no choice but to frame it that way, and act on it.
Even if it does, on net, the transition path itself drains demand. The model is about the dynamics, not the long-run steady state.
Section 5 of the paper rules out upskilling as a sufficient remedy. The externality is between firms, not inside workers' heads.
A levy proportional to displaced labor-hours per automated task, rebated as universal dividends, preserving capability gains while restoring demand.
Translated by a designer who thinks research deserves an interface.
Sunil Kumar R
Sunil reads dense research so you don't have to. He believes the most important AI conversations of this decade, automation, displacement, policy, are too consequential to live behind paywalls and equations. This explainer is part of a series translating frontier work into interfaces ordinary people can actually reason with.
"Design isn't the polish on top of a system. It's how a society decides what's worth understanding."