Something’s broken in the economy right now.
And everyone’s celebrating it.
AI agents are killing friction everywhere. Sounds great, right? Friction is waste. Friction slows things down. We’ve spent centuries eliminating it—better roads, faster shipping, streamlined everything.
Now we’re about to eliminate it completely.
Here’s what nobody wants to say out loud: that friction was the only thing holding this whole thing together.
The Invisible Architecture of Waste
Buy something online today. What happens?
Someone processes your order. Someone manages inventory. Customer service handles your questions. Accountants reconcile payments. Managers review reports.
All these people slow things down. They make mistakes. They take lunch breaks. They’re expensive.
They’re also the reason the economy works.
Every one of those “inefficient” humans takes their paycheck and spends it. Groceries. Rent. Restaurants. Streaming services. That spending creates demand. Which employs more people. Who spend their paychecks. The whole beautiful, messy cycle continues.
AI agents don’t do any of that.
They process transactions in milliseconds. Never sleep. Never make mistakes. Never buy anything.
McKinsey says AI agents will mediate $3-5 trillion in commerce by 2030.[1] Trillion. With a T. They’re talking about it like it’s pure upside—faster transactions, lower costs, better margins.
But where does that $3-5 trillion come from?
It comes from eliminating the humans who used to facilitate those transactions. Eliminate the humans, eliminate their paychecks. Eliminate their paychecks, eliminate their spending. Eliminate their spending…
You see where this goes.
The Speed Problem
Previous automation waves gave us time to adapt.
The industrial revolution displaced agricultural workers over generations. People retrained. New industries emerged. The economy restructured itself—slowly, painfully, but it restructured.
We don’t have generations this time.
We might not even have years.
I’m watching companies deploy AI agents right now. Not pilots. Scale deployments. Customer service departments that had 200 people six months ago? Now they have 50. Procurement teams replaced by autonomous systems that negotiate contracts while everyone sleeps. Marketing operations that needed coordination across multiple teams? Running on autopilot.
The efficiency gains are real. Spectacular, even.
Margins expanding. Stock prices climbing. Executives getting bonuses for “digital transformation.”
And underneath it all, the foundation is crumbling.
The Demand Destruction Nobody’s Modeling
Here’s what keeps me up at night:
Every economic model we use assumes a certain level of employment. Consumer spending models. GDP projections. Inflation calculations. They’re all built on the assumption that most people work and spend their earnings.
What happens when that assumption breaks?
We’re about to find out. And I don’t think we’re ready for the answer.
Citrini Research paints a scenario where this plays out over the next 2-3 years.[2] Speculative? Sure. But the logic is airtight:
2026: Companies deploy AI agents aggressively. Layoffs accelerate. Profits soar. Markets rally.
2027: Unemployment climbs. Consumer spending contracts. The companies that just automated everything suddenly realize their customers can’t afford their products anymore.
2028: Economic crisis. Not because the technology failed. Because it succeeded too well.
The scary part?
We’re already in the first phase. Right now. Today.
The Paradox of Productivity
We’ve always assumed productivity gains are good. More output per worker means higher living standards, right?
That’s been true for most of human history.
But there’s an assumption buried in that logic: the productivity gains get distributed broadly enough to create purchasing power.
When factories automated, factory workers lost jobs, but new jobs emerged in services. When computers automated clerical work, people moved into knowledge work.
This time feels different.
AI agents aren’t just automating tasks. They’re automating entire job categories. Simultaneously. Across the economy.
Customer service? AI agents. Sales? AI agents. Procurement? AI agents. Basic legal work? AI agents. Accounting? AI agents. Marketing operations? AI agents.
Where do all those people go?
“Learn to code” isn’t an answer when AI is writing code. “Move into creative work” isn’t an answer when AI is doing creative work. “Focus on human connection” isn’t an answer when companies realize AI agents are cheaper and customers don’t actually care.
The Concentration Accelerator
Here’s another piece nobody talks about:
AI agents concentrate wealth in ways previous automation didn’t.
When you automated a factory, you still needed workers to run the machines. When you automated clerical work, you still needed people to manage the systems. The productivity gains were shared—however unevenly—between capital and labor.
AI agents change that equation fundamentally.
The gains accrue almost entirely to whoever owns the compute. And compute ownership is extraordinarily concentrated. A handful of companies control the infrastructure that’s about to mediate trillions in economic activity.
Double whammy: Employment drops, reducing distributed purchasing power. Meanwhile, gains from automation concentrate in fewer hands. The people who benefit most from AI agents are the least likely to spend their gains in ways that create broad economic demand.
What Friction Actually Did
I’ve been thinking about what we’re losing when we eliminate economic friction.
It’s more than just inefficiency.
Friction distributed opportunity. All those “unnecessary” intermediaries in supply chains? They were businesses. They employed people. They created opportunities for entrepreneurship and local economic activity. Remove them, remove the distributed network of economic participation.
Friction created buffers. Slow human decision-making meant economic shocks propagated gradually. You had time to adjust, respond, adapt. AI agents operating at machine speed could create cascading failures that move faster than any human institution can respond to.
Friction enforced limits. You couldn’t scale infinitely when humans were in the loop. Natural constraints on how fast you could grow, how much you could automate, how quickly you could disrupt.
Those constraints are gone now.
We’re about to find out if they were load-bearing.
The Transition Nobody’s Planning For
The optimistic scenario: we figure out new economic models fast enough to prevent catastrophe. Universal basic income. Compute taxation. Radical redistribution. New forms of value creation that AI can’t automate.
Maybe.
But look at how long it took us to adapt to much slower technological changes. Look at how our political systems struggle with problems that move at human speed.
Now imagine trying to restructure the entire economic foundation of society in 2-3 years while unemployment is spiking and social stability is fraying.
I’m not optimistic about our ability to manage this transition smoothly.
What This Means for You
If you’re running a business: You’re facing an impossible choice. Deploy AI agents and gain efficiency, or fall behind competitors who do. But if everyone makes the rational individual choice, we collectively destroy the consumer base we all depend on.
Classic tragedy of the commons. Except it’s playing out at the speed of software deployment instead of over generations.
If you’re an employee: The calculus is even worse. Your job might be safe today, but the velocity of change is accelerating. The skills that protect you now might be automated next quarter. And the traditional advice—retrain, upskill, adapt—assumes there are safe harbors to retrain for.
I’m not sure those exist anymore.
If you’re an investor: You’re trying to price assets in an economy that might not resemble anything we’ve seen before. The companies with the best AI agents might have the best margins but the worst long-term prospects if their customer base evaporates.
The Question We’re Not Asking
Everyone’s asking: “How fast can we deploy AI agents?”
Nobody’s asking: “How fast should we deploy AI agents?”
We’re treating this like a technology problem when it’s actually a systems problem. The economy isn’t just a collection of individual transactions to be optimized. It’s a complex adaptive system with feedback loops and emergent properties and tipping points.
We’re about to find out what happens when you optimize one variable—efficiency—without considering what it does to the system as a whole.
My guess? We’re going to learn some very expensive lessons about the difference between local optimization and global stability.
The Uncomfortable Truth
The reduction of economic friction through AI agents is probably inevitable.
The technology exists. The competitive pressures are real. No individual company can afford to fall behind.
But inevitable doesn’t mean painless.
And it doesn’t mean we shouldn’t try to manage the transition better than we’re currently doing.
Right now, we’re sleepwalking into a transformation that could make the Industrial Revolution look gentle by comparison. We’re celebrating productivity gains without modeling the demand destruction. We’re optimizing for efficiency without considering stability.
We’re moving at software speed through changes that require social and political adaptation.
The friction we’re eliminating wasn’t just waste. It was the structure that held the economy together.
And we’re about to find out what happens when you remove load-bearing walls because they’re “inefficient.”
I hope I’m wrong about all of this. I hope new economic models emerge faster than I expect. I hope we find ways to distribute the gains from AI broadly enough to maintain purchasing power. I hope the transition is smoother than the logic suggests it will be.
But hope isn’t a strategy.
And right now, I don’t see anyone with a credible plan for managing what’s coming.
The frictionless economy is arriving.
Whether we’re ready for it or not.
Be the first to comment