I am more anxious right now than I have ever been in my professional career. I am also, and I mean this literally, pee-my-pants excited.
Generally I am not an anxious person. If the Eagles are in the Super Bowl, sure. Business, not so much. I have built and advised companies through the dot-com bubble, the 2008 financial crisis, COVID, and a half-dozen client crises that should have killed somebody’s business. I thrive in chaos. Weirdly, my world slows down in wartime.
So here is why I am anxious. For the first time in nearly three decades of running and advising companies, I cannot find a durable, defendable position for a company to win over time. Not yet.
My pattern is that I am early. If you were at my Leadership Summit in March 2022, you watched me demonstrate AI tools before ChatGPT existed. I am three to five years early on most calls, five to seven on the big ones. The bet I am running on AI says we have somewhere in that window before the math the academic economists are now formalizing shows up in your P&L, in your team, in your industry, and in the demand environment your business sits on. A decade at the outside. It could come faster if the technology keeps compounding and implementation keeps getting easier.
I am writing this because I do not think most of you see the shape of it yet. You are heads down. You are running the business. That is the job. But the ground is moving underneath you, and the position you hold today is not the position you will hold soon, whether you act or not. Knowledge and professional services first. Manufacturing, industrial, and construction after.
The trap
There is a new academic paper making the rounds that articulates what I have been watching since I got my hands on the first AI writing tool in 2021. The authors call it the AI Layoff Trap. I will call it what it is: a race that ends with every runner worse off, where dropping out is the only move that gets you eliminated first.
Here is the argument in operator language.
When you replace a worker with AI, you capture one hundred percent of the cost savings. You bear only a fraction of the demand destruction. Your laid-off worker was also somebody’s customer, and their lost spending shows up across every business in your category, including yours, but mostly your competitors’. So the math each of you runs in isolation looks like a no-brainer. Cut the headcount, pocket the savings, hold the price, make a lot of money.
Except everyone is running the same math at the same time, and the aggregate purchasing power of the people you are collectively laying off is the demand floor your whole industry stands on.
You automate your way to boundless productivity and zero customers.
Why smart people race off the cliff anyway
Here is the part that should make you stop reading and stare at the wall for a minute.
The authors show that automating is a strictly dominant strategy. That is a technical term. It means that no matter what every other firm in your category does, your individually optimal move is still to automate. If they restrain, you take share by automating. If they automate, you have to automate to keep up. There is no version of the game where holding back is your best move.
This is not a coordination failure. Coordination failures can be solved with a meeting. You call your competitors, you agree to slow down, and the agreement holds because cooperation serves everybody once the defecting stops.
This is a prisoner’s dilemma. Even after the meeting, your dominant move is still to defect. If every CEO in your industry signed a blood oath tomorrow to freeze headcount, each one of them, including you, would still be better off breaking it the next morning. The agreement is not self-enforcing, because the underlying incentives did not change.
Two more findings, and both of them are worse than the first. More competition makes it worse, not better. A monopolist would moderate automation because it would eat the full demand loss itself. A fragmented industry, which is what most of you operate in, spreads that cost across all the players, so more competitors means a smaller share of the damage per firm and a faster race to the edge.
And this is the line that stopped me cold: better AI amplifies the problem rather than solving it. The more capable the technology, the bigger the per-task savings, and the wider the gap between what you privately want to do and what your industry collectively should do. The technology is not slowing down.
The early adopters are winning, and it is not close
If you doubt the structural argument, look at the scoreboard.
Nvidia closed fiscal 2026 with roughly 42,000 employees and about $5.14 million in revenue per worker. More than double Apple. Nearly triple Microsoft. Revenue grew 65 percent year over year while headcount grew under 17 percent. Net income landed near $120 billion, about $2.86 million per employee.
A new class of AI-native firms is reaching $500 million in annual revenue with teams under fifty people. Cursor. Midjourney. The metric that used to be aspirational, one million dollars of revenue per employee, is now a floor rather than a ceiling.
It is not only software. Real Brokerage, an AI-first company of roughly 500 employees and a much larger independent agent base, merged with Re/Max to create the second-largest real estate company in the world. Real’s CEO leads the combined company. Real’s side owns 59 percent of it. Headquarters moves from Denver to Miami. That is a wild set of facts for anyone who thinks this stays in tech.
These firms are not struggling. They are pulling away in capital, in compute, and in capability, far enough that the rest of the field needs a different category of strategy to compete at all.
And the layoffs have started
Now look at the other side of the ledger.
Block cut close to half of its ten thousand people in February and published the reasoning in From Hierarchy to Intelligence. Its CEO said the quiet part out loud: most companies will reach the same conclusion inside a year. Salesforce replaced four thousand customer support agents with agentic AI. Atlassian announced ten percent layoffs while its stock fell 54 percent in four months on AI disruption fears. Goldman Sachs is piloting an autonomous coder that lets one senior engineer do the work of a five-person team. The US economy shed 92,000 jobs in February, the largest single-month loss in years.
US job postings are down roughly 32 percent since ChatGPT launched. AI-related postings are up 92 percent this year. The labor market is not weakening. It is splitting. Two different economies are coming out of the same data set, which is the K I laid out in the 2026 Economic Outlook. College graduates are booing commencement speakers who bring up AI. I understand the impulse.
What does not work
Before I get to what does, here is the list of fixes the paper rules out. Most of them are in your feed right now, proposed by serious people.
- Universal basic income. Raises the floor on living standards. Does not change a single CEO’s automation calculus. The math at your desk is untouched. I argued for this in 2025 and the Age of AI, and I was solving the wrong equation.
- Capital gains or profits taxes. They scale down your post-tax dollar without moving the marginal automation decision. You automate at the same rate.
- Worker equity participation. Helps. Narrows the gap. Cannot close it unless workers own more than one hundred percent of the company.
- Voluntary agreements between firms. Cannot work, for the reason above. The oath is not self-enforcing.
- Upskilling and retraining. Helps at the margin by recycling some displaced income back into demand. Does not eliminate the trap.
- Wage adjustment. Wages fall, automation slows. But the only wage level that closes the trap is roughly the cost of the AI itself, which means the people who still have jobs are earning machine wages. The trap gets solved by impoverishment.
The one mechanism the authors find that actually works is a per-task automation tax set equal to the demand destruction the average firm pushes onto its peers. A Pigouvian tax, the same instrument economists use for pollution. It changes the decision at your desk, which is the only place the decision is made.
That tax does not exist. There is no political coalition currently capable of passing it in any jurisdiction that matters, and I put the odds of it arriving before this plays out as low. Government moves in decades. This is moving in milliseconds.
What this means for you
So here is the position you are sitting in. You see the cliff. The authors see the cliff. Your competitors see the cliff. And the paper proves that seeing it does not help. Foresight is not the bottleneck. Rationality is not the bottleneck. Goodwill is not the bottleneck. Structure is the bottleneck, and from where I sit the structure is not currently correctable.
I think that view is shared in places you would not expect. Anthropic has shipped products that visibly moved stock prices in cybersecurity, legal services, and software, and its CEO has said publicly that he expects software to become essentially free. That is a founder describing the commoditization of his own customers’ businesses, out loud, on a stage.
Capitalism demands efficiency. The system rewards the firm that automates fastest and punishes the one that holds back. Restraint here is suicide by virtue. The CEO who refuses to automate on principle does not save the system. They exit it first, and their competitors finish the job anyway.
This is the point where every founder I know wants me to produce a clever third path. There isn’t one. Not at the level of one CEO, and not at the level of one industry. But there is a position to hold.
The operating answer
You participate. For as long as possible. With your eyes open. That is not cynicism, it is the only logically consistent move once you accept the structure. Here is what I mean by it.
1. Automate with the contraction in mind, not the peak
Most of your peers are building AI stacks that assume the demand environment stays roughly stable. It will not. Build the cost structure that survives the contraction, not the one that maximizes margin at the top. Variable beats fixed. Modular beats integrated. Reversible beats sunk.
2. Stop competing on the same automation curve
If everyone in your category is racing to replace the same headcount with the same tools, the savings get competed away inside eighteen months. The price resets. You end up with the margin you had before, minus the customers. Find the work AI cannot do in your category and build the moat there. In most industries that is relationship, judgment, trust, accountability, and proprietary domain knowledge. None of those are automated yet.
3. Get bigger and freer, faster
The companies that come through the contraction will be the ones with the strongest balance sheets, the lowest founder dependency, and the most resilient demand base. The founder bottleneck is not a coaching cliché in this context, it is a survival variable. A founder-dependent company is the most fragile asset class there is in a demand shock, because you cannot raise, cannot sell, and cannot pivot quickly when the founder is the constraint on every decision. If you want the honest read on where yours stands, take the scale readiness assessment.
4. Watch the income replacement rate
That is the academic term. In plain English: how fast are displaced workers getting back to comparable wages? When that number is high, the trap weakens. When it is low, which is where we are, the trap tightens. It is your leading indicator for the speed of the contraction. Watch it the way you watch your cash conversion cycle.
5. Do not pretend the trap is not there
Every CEO I have talked to about this in the last sixty days reacts the same way: a flash of recognition, then a fast retreat to “but my situation is different.” Your situation is not different. The math is the math. The honest move is to name what you are participating in, decide the terms of your participation, and stop telling yourself a story about creating the jobs of the future. You might be. The aggregate evidence says most of us are not.
6. Assume very few companies are left standing, and go deep
There will be companies standing. There will be fewer of them than you think, and the concentration of wealth will be astronomical. Block’s framing is the one I would use. A system can now hold a continuously updated model of an entire business and coordinate work that used to require humans relaying information up and down layers of management. What it cannot do is touch the world. As Block puts it, “A world model that can’t touch the world is just a database.” People at the edge sense what the model cannot: intuition, cultural context, trust dynamics, the feeling in a room, and the ethical calls that should not be made by a machine. The edge survives. The layers between the edge and the model do not.
Which means you have to go deep in something complex and hard to understand. Most of us do not currently live in that world. That is the work.
The close
I am not writing this to talk you out of automation. I am writing it because the most dangerous CEO in this environment is the one who believes they are making an isolated, rational, value-creating decision. You are not. You are one player in a multi-player game whose equilibrium is collectively destructive and individually unavoidable.
The race is on and you are in it. Pretending otherwise costs you the company.
So the work is to run it in a way that leaves you standing when the music stops. Build for the contraction. Lower your founder dependency. Strengthen the demand base you actually control, which is your existing customers, your relationships, and your reputation. Hold cash. Move with clarity instead of panic.
The race is not the one you thought you were in. It was never “who builds the best business.” It is “who is still here when the demand floor cracks.”
Win that one.
Jerry
PS. For those of you about to accuse me of arm-waving with no answer, you are right. And when you tell me the world has always evolved forward through every advance, you are right about that too. I just cannot get to a place yet where new thinking or new innovation is not almost immediately absorbed into AI and commoditized, and that leads me to question how value gets created and exchanged at all in the future. I would love for you to show me a different view. If you have one, put it in front of me.