Most of the AI-and-jobs conversation runs on a simple story: the machines get good enough, they do your work for less, and you’re out. It’s a story about capability — about the day the model can finally do the task. If that were the whole picture, the preparedness answer would be equally simple: stay ahead of the curve, keep your skills sharp, be the person who runs the AI instead of the person it replaces.
I’ve repeated a version of that advice myself. I still think it’s sound. But a paper I came across recently convinced me it’s incomplete in a way that matters for how you plan. The threat isn’t only that AI can replace workers. It’s that the structure of a competitive market can push companies to replace them faster and further than is good for anyone — faster than is good for workers, for the broader economy, and, strikingly, for the companies doing the firing.
That’s a different kind of problem. You can out-train a capability threat. You can’t out-train a structural one. So it’s worth understanding exactly what these two economists are claiming, where it’s solid, and where it’s still just a model on a page.
What the paper actually argues
The paper is “The AI Layoff Trap” by Brett Hemenway Falk and Gerry Tsoukalas, posted to arXiv in March 2026 and revised that June. It sits in theoretical economics, which is the first thing to be honest about: this is a mathematical model of how firms behave, not a measurement of layoffs that have already happened. It’s an argument about incentives, and it has not completed formal peer review that I’m aware of.
Here’s the mechanism, in plain terms.
When a company automates a job away, it captures the full savings from that cut — the whole salary, right to its own bottom line. But automation has a second effect that’s easy to miss: the fired worker is also a customer, somewhere in the economy. Their lost wages mean lost spending. That drop in demand doesn’t land on the company that did the firing. It gets spread across the whole market — including all of that company’s competitors.
So each firm faces a lopsided deal. All of the savings from automating come home to you. Only a fraction of the demand damage does; the rest is somebody else’s problem. Every firm sees the same lopsided deal, so every firm automates harder than it would if it had to swallow the full cost of the demand it was destroying. The authors call the result an automation arms race.
Each company fires its workers and keeps the savings. The lost spending those workers represented gets quietly billed to everyone. Multiply that across an economy and you get more layoffs than even the shareholders would choose — if they could choose together.
Economists have a name for this shape of problem: a demand externality. It’s the same logic as pollution. A factory that dumps waste in a river gets all the benefit of not paying for cleanup, while the cost flows downstream to everyone else, so it pollutes more than is socially optimal. Here the “pollution” is lost consumer demand, and the river is the economy we all sell into. The paper’s unsettling claim is that the firms end up collectively worse off — they automate past the point that maximizes even their own aggregate profits, because no single firm has a reason to stop first.
Why this is the scarier version of the story
Sit with the implication for a second, because it inverts the comfortable advice.
If AI job loss were purely about capability, it would arrive unevenly and slowly — task by task, as the technology matured — and the people who adapted would be fine. The market would find a new equilibrium. That’s the version where “learn to use the tools” is a complete answer.
The trap version is worse in three specific ways:
- It doesn’t wait for a downturn. This dynamic runs in a healthy, competitive, profitable market. The layoffs aren’t a symptom of a company struggling; they’re a symptom of the company competing. You can’t read the usual warning signs — falling revenue, a shaky sector — because they may not be there.
- It overshoots. The model’s whole point is that the economy blows past the efficient amount of automation. More jobs go, and go sooner, than the underlying technology alone would justify. Individual prudence doesn’t throttle it, because the pressure is on the firms, not on you.
- The obvious fixes don’t fix it. This is the part that stopped me. The authors test the standard menu of remedies and find most of them don’t eliminate the problem — not capital-income taxes, not giving workers equity in their employers, not universal basic income, not upskilling programs, not even direct bargaining between the parties. In their model, only one intervention closes the gap: a Pigouvian tax on automation — a tax that makes a firm pay for the demand destruction it causes, the same way a carbon tax makes a polluter pay for emissions.
I want to be careful here. “UBI doesn’t work” is a strong-sounding claim, and in the paper it means something narrow: UBI doesn’t stop firms from over-automating in this specific model. It might still be a perfectly good way to support displaced people — that’s a different question the model isn’t answering, and it’s not something I’m willing to tackle here. Don’t let a headline flatten that distinction.
The honest caveats
A model is a claim about how the world works, and this one is young. Three things keep me from treating it as settled:
- It’s theory, not a body count. The paper shows that if firms behave the way the model assumes, over-automation follows. It is not evidence that this is what’s driving the AI layoffs we’re already seeing. Reality has friction the model doesn’t.
- It hasn’t cleared peer review. Smart economists will poke at the assumptions — how substitutable labor and automation really are, whether demand leaks across the whole market as cleanly as modeled, whether firms are as myopic as the setup requires. Some of those punches may land.
- The remedy is a policy fight, not a plan you can enact. A Pigouvian automation tax is a genuinely contested idea with real trade-offs — tax automation too hard and you kneecap the productivity growth that raises everyone’s living standards. Whether it ever happens is a political question far outside your kitchen.
None of that makes the paper wrong. It makes it a well-formed warning rather than a verified forecast. For a prepper, a credible warning about a mechanism is exactly the kind of thing worth acting on before the body count exists — that’s the whole discipline.
Why this resonated with me today
This paper resonated with me for one specific reason — I’m seeing signs of it within my current employer. Over the past 6 months or so, I’ve noticed a dramatic uptick in capital spending in the automation and Industry 4.0 spaces. I think most of you are familiar with automation — robotics, vision systems, etc. They’ve been around for a very long time (I was working with robots 25 years ago). Industry 4.0 is a more recent concept and refers to a “fourth industrial revolution” that is currently underway. There are several technologies that are driving this shift:
- Industrial Internet of Things (IIoT): Sensors and equipment that collect and share real-time operational data. This is normally collected via SCADA (Supervisory Control and Data Acquisition) or other data warehouses. Using this data, the performance of processes and equipment can be monitored and adjusted in real-time.
- Cyber-Physical Systems / Digital Twins: Every physical object within a process has a virtual counterpart (the digital twin). This allows software to model processes to run simulations, predict failures, etc. in a virtual space. It allows changes to be simulated before you ever make a physical change, but it also lays the groundwork for implementing robotics and other automation technologies as the digital twin can be used to train the automation.
- Artificial Intelligence (AI) and Machine Learning: Specifically, when combined with cameras for detection of quality defects, ergonomic issues, safety issues, etc. that just a year ago required people to detect.
Over these months, I’ve wondered why my company was making this push now. Everywhere I’ve worked has always pursued automation, but only when it made economic sense (the payback justified the capital expenditure). What I’m seeing now is that projects that would have never been considered based on the payback are now the top of our priorities. I had wondered why the sudden push, and then I came across this paper. It definitely made me pause and consider if my current employer is falling into this trap.
While correlation doesn’t automatically mean causation, I also believe in Gibb’s Rule #39: There are no coincidences.
What it changes for how you prepare
Here’s the practical translation, and it’s less about AI than you’d think.
Go back to the Job Loss threat. It’s already the single most likely serious disruption your household will face — not nuclear war, not a grid-down scenario, but a layoff. The AI-layoff-trap argument doesn’t invent a new threat. It does two things to the one that was already at the top of your list:
It raises the Likelihood, and uncouples it from the economy’s health. If this dynamic is real, a layoff can arrive while your employer is profitable and your industry looks fine. The reflex of “I’ll see it coming because the company will be struggling” is exactly the reflex the trap defeats. Plan as if the warning may not come.
It argues for resilience that doesn’t depend on your specific job surviving. You can’t out-skill a structural pressure on your whole occupation, so the hedge can’t be “become irreplaceable.” It has to be the boring, proven stuff that makes an income gap survivable no matter why it happened:
- A cash reserve deep enough to cover several months of essential expenses — the one variable that most determines whether a layoff is a bad quarter or a lost house.
- A bare-bones budget you’ve already worked out, so you know your real floor before you need it.
- Low high-interest debt, so a setback doesn’t compound. (Our Debt Payoff Planner exists for exactly this.)
- A deep pantry so food spending can flex toward zero for a while.
- Skills and a network kept current — still worth it, just no longer the whole plan.
That’s not a new prescription. It’s the Job Loss playbook you should already be running. The paper’s contribution is to take away the excuse that it won’t happen to you because your company is doing well. In a market that over-automates, doing well might be the problem.
Bottom Line
- The threat isn’t just capability — it’s incentives. “The AI Layoff Trap” argues that competitive firms are structurally pushed to automate past the point that’s good for workers, the economy, or even themselves, because each firm keeps all the savings but shares the demand damage.
- It’s a model, not a measurement. The mechanism is a well-formed economic argument that hasn’t cleared peer review and isn’t proof of what’s driving current layoffs. Treat it as a credible warning, not a forecast.
- The comfortable fixes come up short in the model. Upskilling, UBI, worker equity, and capital taxes don’t stop the over-automation in the authors’ math — only a tax on automation does. That’s a policy fight you don’t control, so don’t wait on it.
- It supercharges a threat you already rank first. Job Loss is the likeliest serious disruption most households face. This argument raises its Likelihood and detaches it from the usual warning signs. The hedge is the same as always — cash reserve, lean budget, low debt, deep pantry — you just have less reason to assume you’ll see it coming.
Rate it for yourself on the Job Loss threat page, and if your reserve isn’t where it should be, start with the Debt Payoff Planner.