It is not mainly a story about jobs. It is a story about the price of thinking, the price of money, and the executives who will see the coming disruption and prepare for it now.
The dominant narrative around AI and employment is binary: jobs survive, or they vanish. Optimists point to every prior technology wave and argue that new work always emerges. Pessimists paint apocalyptic pictures of mass unemployment. Both framings miss the more immediate and more dangerous dynamic that is already underway, and both miss it for the same reason; they are looking at headcount when they should be looking at price.
AI is not primarily destroying jobs. It is repricing them. The cost of producing competent cognitive work is collapsing, and as it falls, the scarcity premium that knowledge workers have commanded for decades collapses with it. That is the first front of what we call the Great Repricing, and it is the one you can already measure. But it does not stop at labor. The same force that reprices the work eventually reaches the money, because the political response to displacement gets paid for, and how it gets paid for reprices the dollar itself. Labor first, money next; two fronts of one phenomenon.
There is a number underneath all of this that executives should sit with before anything else. The cost of a unit of machine cognition has fallen roughly 280 times in about two years, with the median rate of decline running near fiftyfold per year. That is not a normal technology cost curve; it is closer to the early semiconductor era compressed into months. When the price of an input falls that fast, the question is never whether the things built on that input get repriced. It is only how fast, in what order, and who is positioned when it happens.
And here is the part that matters most for the company you run: the executives who recognize this shift and rush to act on it will, in most cases, fare worse than the ones who recognize it and act deliberately. The repricing is real. The panic it is generating is not a strategy.
The scarcity premium that built the white-collar economy
For most of the last half-century, knowledge workers enjoyed a structural advantage in the labor market. Their expertise was hard to acquire, slow to develop, and expensive to replace. A company that lost a senior financial analyst or a mid-career software engineer faced months of recruiting, onboarding, and ramp-up before the replacement reached full productivity. That friction gave workers leverage; it kept wages rising, and it made employers think twice before issuing ultimatums.
This was not unique to any single profession; it was the foundational logic of the entire white-collar economy. From consulting to legal services to software engineering to corporate finance, the value proposition was the same: you need what I know, and you cannot easily get it elsewhere.
AI has not eliminated the need for knowledge work. What it has done is dramatically expand the supply of cognitive output available to any organization with a subscription and a willingness to redesign its workflows. When a model can produce an eighty-percent-quality first draft of a legal brief, a financial analysis, or a marketing strategy in seconds, the marginal value of a human producing that same draft falls. The worker is not replaced; the work is worth less. That distinction matters enormously, because repricing looks nothing like displacement in the headlines, and it can be just as consequential over time.
Front one · Labor reprices first
The repricing of labor is not theoretical, and it is not a forecast. The early evidence is already striking.
White-collar payrolls in the United States have contracted for more than two consecutive years, a streak that, according to former Glassdoor chief economist Aaron Terrazas, is essentially unprecedented outside a recession going back seven or eight decades. Yet the headline unemployment rate has hovered around 4.3 to 4.4 percent, which masks how concentrated the pain actually is. The aggregate number looks calm because the damage is not spread evenly; it is pooling in specific roles and specific cohorts.
Companies are mostly not announcing dramatic layoffs. They are doing something quieter and harder to see: not backfilling roles when people leave, freezing headcount, stretching existing teams thinner, and filling the gap with AI tools. Hiring in consulting has fallen on the order of forty percent from its peak. Placement rates at even elite MBA programs have deteriorated sharply; Duke’s Fuqua School reported that roughly a fifth of graduates were still seeking work three months after graduation, up from about five percent in 2019. Goldman Sachs Research has called AI the big labor story of 2026, noting that job growth slowed meaningfully in the second half of 2025, and the Federal Reserve has acknowledged that private-sector hiring has effectively stalled.
Who loses leverage, and who gains it
The repricing is selective, and the line it draws is the most useful piece of analysis in this entire discussion. Research published by the Dallas Federal Reserve in early 2026 separates two kinds of knowledge: codifiable knowledge, the textbook learning and established procedures a model can absorb, and tacit knowledge, the judgment that only accumulates through experience. The distinction turns out to be decisive. For occupations with high experience premiums, AI exposure is associated with positive wage growth; the technology complements seasoned workers by automating the routine components of their work. For occupations with low experience premiums, particularly entry-level roles, AI exposure correlates with meaningful wage decline.
In plain language: if your value comes from knowing things a model can also know, your leverage is evaporating. If your value comes from judgment that only experience can build, you may actually benefit, at least for now. None of this registers as a crisis in the macro data. It shows up as a slow, grinding deterioration that workers feel and few can name. The repricing is real, it is already here, and it is accelerating.
The bridge · A demand crisis hiding inside the efficiency story
If the repricing only touched individual workers, it would be painful but manageable. The deeper risk emerges when you follow the logic to its collective conclusion.
The CEO who trims her workforce and resets wages downward is making a rational decision for her company this quarter. But if every CEO in every industry reaches the same conclusion at roughly the same time, the aggregate result is not efficiency; it is a demand problem. This is the Henry Ford insight run in reverse. Ford grasped that his workers were also his customers, and that paying them enough to buy the cars they built was not generosity but market creation. The repricing threatens to unwind that logic across the knowledge economy all at once.
The mechanism is a wage cascade. Displaced white-collar workers do not vanish; they step down the ladder. The former project manager drives for a rideshare service; the former analyst competes for lower-paid service work. That floods the labor-intensive sectors with educated workers and compresses wages there too, so the problem widens from “professionals are under pressure” to “income expectations across the upper half of the consumption stack are deteriorating.” The Citrini Research essay by Xupeng Chen that popularized the term “Ghost GDP” traces exactly this path: output keeps rising and corporate profits look strong, but the income stops circulating through households. A recent academic working paper out of NYU models the same wedge formally, describing how AI-generated output can substitute for labor-generated output while the income that would have flowed to those workers simply goes missing from the consumption economy. When several independent analysts and a peer-style paper are converging on the same dynamic, it is no longer a fringe worry; it is a thesis worth pricing into your revenue assumptions.
This is especially relevant if your customers are knowledge workers, or companies that employ them. The efficiency you gain by reducing your own headcount can be offset by the revenue you lose selling into a customer base whose purchasing power is eroding. The demand problem is not a reason to ignore the repricing. It is a reason to approach it with far more care than the current climate of urgency suggests. It is also only half of the macro picture, because a demand crisis of this kind does not sit still; it provokes a response.
Front two · The Debasement
The repricing of work does not happen in a vacuum. It happens inside a political economy that will not tolerate the displacement quietly. As white-collar work compresses, the most politically influential constituency in any developed society, the educated professional class, comes under direct economic threat for the first time in living memory. A government response is not a possibility; it is a near-certainty. What remains uncertain is its shape.
There are utopian claims that AI will enable society to provide citizens their basic needs without the need for them to work at all, but that can only come to pass (if it ever does) after a gauntlet of supply chain and delivery chain re-engineering across every industry. The unemployed lawyers and programmers will demand a response far sooner than all of those dynamics can play out. So it’s clear that the government response will take the form of transfers: expanded safety nets, and something with the shape of a universal “UBI” stipend. The pivotal question is how those transfers get funded, because the funding mechanism, not the transfers themselves, is what determines whether the dollar gets repriced.
There are two roads. The transfers could be paid for by taxing the AI windfall directly, which redistributes wealth without debasing the currency, and that is the path some of the technology’s own architects advocate. While there may be some kind of tax program created to appease the anger of the voting public, we do not believe that will be much more than lip service; it cannot be set at a scale which can fully fund nationwide UBI, for four reasons:
- A capitalist system cannot coherently punish the innovation it is simultaneously trying to win, as long as the US cares anything about retaining our place as the top economy in the world.
- The paper trail required to tax AI-generated value at the necessary scale is far harder to construct and enforce than its proponents assume. Even today, with no financial penalty involved, people and organizations hide their use of AI and view the details as proprietary secrets.
- Money printing policy can be decided at a conference room table by a group of decision makers; it does not require passing of legislation by politicians with divided agendas.
- The political pressure to remove people’s misery will completely overwhelm the academic desire to manage the money supply.
So in the United States specifically, against a backdrop of runaway national debt, the path of least resistance runs through the printing press, not the tax code. We admit that this is the load-bearing assumption in our thesis, and it is the one we watch most closely. And this creates a unique dynamic: the currency debases (inflationary) even as cognition deflates in price. This is the mechanism investment analyst Lyn Alden calls “fiscal dominance”: structural deficits that force ongoing currency debasement regardless of who holds office. To her theory we can now name the specific force on the other side of the equation. Alden identifies technology, debt, and demographics as the major deflationary pressures in the abstract; we think AI is about to make the “technology” term dominant and abrupt. Crucially, the displacement it causes is itself the political trigger for the transfer spending that drives the debasement. The same force sits on both sides of the ledger. AI is both the deflationary shock and the cause of the inflationary response.
The collision, and why it is receiving too little attention
Put the two fronts together and you get something stranger than either alone. The price of thinking falls while the price of money falls too. Labor and the dollar get repriced at the same time, in opposite directions, and the turbulence lives in the collision.
The two forces do not cancel; they hit different baskets. Anything intensive in cognitive labor, software, content, analysis, routine professional services, deflates as the cost of producing it collapses. Anything genuinely scarce or physically constrained, housing, energy, skilled trades, the goods and services AI cannot conjure, inflates, and inflates faster when transfer money chases it. The result is not a tidy move in a single direction. It is a relative-price revolution, in which the headline inflation figure can look almost calm while the composition underneath it shifts violently. Averages will mislead. The cost of a knowledge-work deliverable and the cost of a roof over your head will move in opposite directions, and the gap between them is where households, and companies, get sorted.
We want to be precise about the confidence here, because a pillar that overclaims is worth less than one that calibrates. The first front, the repricing of labor, is close to a certainty; it is microeconomics riding a cost curve that has run hard for five years. The second front, the Debasement, is a chain of contingent judgments about how political economy responds, and it rests on the funding assumption named above. We hold the first with conviction and the second as our base case while we watch the fork. What we are confident of is the structure: this is a two-front repricing, and modeling only the deflationary half, as most of the labor commentary does, or only the monetary half, as most of the macro commentary does, gets the picture wrong. Almost nobody is putting both fronts in front of the people who actually have to make operating decisions. That is the gap we are trying to fill.
Move first on capability, not on cuts
This is where the conversation has to shift, because the prevailing response to the repricing confuses two very different first moves, and only one of them is a trap.
In early 2026, Block’s Jack Dorsey cut roughly forty percent of his workforce, taking headcount from over ten thousand to under six thousand. He tied the decision directly to AI-driven productivity, said the business was strong, and predicted that within a year most companies would reach the same conclusion. Block’s stock jumped sharply the next day, the business press treated it as bold leadership, and the implicit message to every other CEO was unmistakable: cut now, or get left behind.
That message is dangerous, but not for the reason the “fast follower” crowd thinks. The mistake is not moving early. The mistake is moving early on the wrong axis. There are two ways to be a first mover in the repricing, and they have opposite risk profiles.
Being first to cut is the trap. Headcount reduction ahead of genuine workflow redesign is an irreversible move made on incomplete information; once the institutional knowledge walks out the door, you cannot call it back, and you will not know what you lost until the contractor invoices arrive eighteen months later. Block, Klarna, and similar firms are technology companies cutting technology workers; their products are digital, their workflows were already software-mediated, and their customer relationships are largely transactional. They operate in the one sector where the distance between intention and execution is shortest. A two-hundred-million-dollar manufacturer, a regional healthcare system, a professional-services firm, or a mid-market financial institution runs on completely different knowledge economics. In a Fortune 500, any individual’s departure is a rounding error; in a mid-market company, losing three or four key people in the wrong departments can mean operational paralysis. The dare-to-quit logic that works for a digital platform can be catastrophic when the person who walks out is the only one who understands your ERP configuration or owns your largest client relationship.
Being first to build capability is the opposite; it is the move that wins, and waiting on it is the real loser’s game. The companies that come through the repricing ahead are the ones already mapping their knowledge topology, redesigning their workflows, accumulating proprietary data, and training their people, while their competitors are still debating whether AI is real. That head start compounds into organizational muscle, data assets, and talent that a latecomer cannot buy back in a single budget cycle, as we have argued at length elsewhere. The first mover who builds aggressively is not absorbing the cost of discovering what does not work; that framing only applies to the company going it alone. The first mover who pairs internal ambition with practitioners who have already navigated AI transformation across multiple industries inherits the lessons instead of paying tuition for them, and captures the full prize rather than splitting it with a fast follower.
So the urgency cuts in two directions at once, and holding both is the whole point. Move slowly and deliberately on irreversible workforce decisions, because those reward patience and punish panic. Move fast and now on capability, because that rewards the early and punishes the late, asymmetrically and without mercy. The executive who reverses these, cutting fast and building slow, gets the single worst outcome available: the institutional damage of the panicked first mover and the competitive disadvantage of the capability laggard. There is no version of this where standing still is the safe choice.
The counterarguments, honestly assessed
We hold this thesis with conviction, but conviction without engaging the best objections is just noise. Four counterarguments deserve real weight.
“Technology always creates more jobs than it destroys.” Historically true, and it may prove true again; agriculture fell from seventy percent of American employment to two percent, and the economy built entirely new categories of work to absorb the surplus. The question is not whether new jobs eventually emerge. It is whether the transition speed of AI displacement outpaces the economy’s ability to create and fill the replacements. Every prior wave operated on a timescale of decades. This one is operating on a timescale of months, and timescale is the whole argument.
“AI augments more than it automates.” Also partly true, and the Dallas Fed data supports it for experienced workers in high-tacit-knowledge roles. But augmentation and repricing are not mutually exclusive. A worker augmented to produce three times the output is a win for that worker and that company; it also means the company needs roughly a third of the headcount for the same output. Augmentation at the individual level is displacement at the aggregate level.
“Companies are AI-washing their layoffs.” The Brookings Institution has argued persuasively that many AI-attributed layoffs are really corrections for pandemic-era over-hiring. That is a fair caution against overstating today’s impact, and we take it seriously. But it is a statement about the present, not a forecast, and model capability is compounding fast enough that extrapolating from yesterday’s data badly underestimates the planning horizon that matters. AI-washed layoffs today simply starts the digging of this hole ahead of the truly AI-driven layoffs.
“The real problem is power, not technology.” The Economic Policy Institute makes a thoughtful case that institutional shifts, union decline, eroded minimum wages, macroeconomic policy, have suppressed wages more than any technology has. That framing has genuine merit for explaining the last forty years. It is less convincing as AI begins to demonstrate something no prior technology could: cognitive work at near-human quality across a vast range of professional tasks, at a marginal cost approaching zero.
None of these eliminates the repricing. At best they suggest the timeline may be longer and the eventual landing zone may include forms of employment we cannot yet picture. At worst they are the same reassurances offered before every prior disruption, recycled for a technology that operates at a fundamentally different speed and scope.
The strategic question for mid-market leadership
The repricing is real. The urgency to cut is manufactured; the urgency to build is not. So the strategic question is not “should we act?” It is “how do we act without becoming the pioneer who absorbs the cost of discovering what does not work?” Five disciplines separate the deliberate from the reactive.
1. Map your knowledge topology before you map your headcount
Understand where your organization’s value actually lives on the codifiable-to-tacit spectrum. The roles most exposed to repricing are those whose primary output is synthesizing, formatting, or communicating information a model can approximate. The roles most durable are those requiring judgment born of experience, relationship trust, and domain intuition no model can replicate. Most mid-market companies have never done this analysis, and without that map any restructuring is guesswork; guesswork with people’s livelihoods is not a strategy.
2. Redesign workflows before you reduce headcount
The most dangerous version of this transition is the one where a company cuts before it has genuinely integrated AI into how the work gets done. Buying subscriptions is not integration. Letting employees use a chatbot for ad hoc tasks is not transformation. Real integration means redesigned processes, retrained teams, validated outputs, and fallback plans for when the models fail. Until that infrastructure exists, aggressive headcount reduction is not efficiency; it is hope disguised as strategy. The companies that win this treat 2026 and 2027 as the assessment-and-redesign phase, not the cutting phase, learning from the pioneers what works at operational scale and preparing to implement it without the restructuring charges and the lost institutional memory.
3. Price your strategy against the whole economy, not just the technology
This is the discipline that distinguishes a genuine AI strategy from a technology roadmap, and it is the reason our counsel tends to sound different from a vendor’s. When we help a company set its IT and AI direction, we are pricing it against the whole-economy backdrop, the labor repricing and the monetary debasement both, not just the capability curve. A plan that is right about the technology and wrong about the economy it lands in is still the wrong plan. If your three-year strategy assumes stable wage costs, stable input prices, and a stable currency, it is quietly betting against everything laid out above.
4. Plan for the demand consequences
If your customers include knowledge workers, or companies that employ them, build the repricing into your revenue assumptions. A fifteen-percent cut in your labor costs means little if your addressable market is contracting by twenty percent because your customers’ customers are under the same pressure. The math has to close on both sides of the ledger.
5. Resist the temptation to benchmark against tech companies
The visible cutters operate where the product is digital, the workflows are already software-mediated, and the customer relationships are transactional. Importing that playbook without translating it to your context is how mid-market companies destroy themselves. The right benchmark is not “what did the platform CEO do?” It is “what is the company most like mine, in my industry, doing successfully?”
The uncomfortable truth
The biggest risk AI poses to the economy is not that it will erase millions of jobs overnight. It is quieter and more corrosive than that. It will strip away the leverage that let tens of millions of knowledge workers command premium pay for their expertise; and then, as the political system moves to cushion that displacement, it will debase the very money those workers are paid in. The jobs may survive in name. The salaries, the benefits, the bargaining power, and the purchasing power attached to them may not. That is the Great Repricing, both fronts of it, and it is already underway.
The biggest risk to your company specifically is different, and more controllable. It is not that you will miss the repricing; the evidence is loud enough now that few leaders will. It is that you will see it clearly, panic, and cut on the wrong axis, shedding people and institutional knowledge before you have built the capability to operate without them, while a more deliberate competitor builds faster than you do and takes the ground you vacate. The companies cutting hardest right now are absorbing the cost of discovering what does not work; the ones building capability with experienced guidance are inheriting those lessons instead of paying for them.
So the instruction is not to wait, and it is not to charge. It is to separate the two moves and get the speed of each one right. Be patient and surgical with the irreversible decisions, the headcount, the people who carry knowledge no model has yet absorbed. Be fast and relentless with the reversible ones, the workflow redesign, the data assets, the organizational muscle that compounds while your competitors are still debating whether any of this is real. You do not need to be the one who cuts first. You very much need to be the one who builds first.
This is also why our counsel tends to sound different from a technology roadmap. We do not price a company’s AI strategy against the capability curve alone; we price it against the whole economy that strategy has to survive in, the labor repricing and the monetary debasement both. The disagreement among the experts about where all of this finally lands, which we mapped across the full optimist-to-pessimist spectrum of AI predictions, is not a reason to wait it out. It is the map of the terrain you have to cross either way. The window for that work is not next year’s offsite. It is now, because repricing rewards the early and punishes the late, asymmetrically and without mercy. The companies that treat this moment as an occasion for clear-eyed preparation rather than panicked restructuring will not merely survive the transition; they will define the next era of competitive advantage.
You cannot choose whether the repricing comes. You can only choose whether it finds you prepared.
