The Great Repricing of Human Intelligence: We Finally Have the Receipts
For five years, we've debated what artificial intelligence would do to employment. Now the numbers are arriving. The early picture isn't mass unemployment or business as usual. It's something considerably more interesting: a fundamental change in what human work is worth.
Well, Here We Are
For roughly five years, we've had the same conversation. Will AI replace jobs? Will it create new jobs? Will productivity explode? What happens when machines can perform the cognitive labor millions of people depend on?
The answers have ranged from utopia to apocalypse, depending on who's talking and how many times they've watched The Terminator.
I've been in the accelerationist camp. Build it. Improve it. Use it. Figure out what becomes possible. That conversation was necessarily speculative. We could demonstrate capability, extrapolate trends, and examine the incentives, but we couldn't point to much evidence of AI reshaping the labor market at scale.
That's changing.
We now have employment data, hiring patterns, productivity measurements, wage premiums, corporate restructurings, and government workforce plans. Together they're answering a better question than whether AI will take our jobs.
The market is starting to reprice human intelligence.
Not human worth. Not human dignity. The economic value of particular kinds of cognitive labor.
The Announcements Are Getting Hard to Ignore
On October 26, Microsoft offered voluntary buyouts to roughly 60,000 employees, about 27% of its 223,000-person workforce. That's an eligibility figure, not 60,000 confirmed departures.
But still.
The same company is forecasting roughly $175 billion in capital expenditures for 2026, substantially tied to AI infrastructure. Microsoft isn't running out of things to build. It's reorganizing what it uses to build them.
That same day, the Dutch government announced plans to cut about 25,000 civil service positions by 2035, targeting €3.5 billion in annual savings. AI is part of the strategy, alongside IT standardization, regulatory simplification, and fewer government activities.
Nobody is claiming 25,000 Dutch civil servants have already been replaced by agents. Nobody is claiming all 60,000 Microsoft employees are redundant.
But both announcements are decisions about human labor made in the context of rapidly improving machines. Real budgets. Real people.
The First Signal: The Jobs That Aren't Being Created
In August, researchers at Stanford's Digital Economy Lab published an updated analysis using payroll data covering millions of American workers.
Among workers ages 22 to 25 in highly AI-exposed occupations, employment was approximately 19% below where it would have been had it kept pace with less-exposed occupations.
That gap was about 15% using July 2025 data.
Here's the interesting part. The divergence showed up primarily as reduced hiring, not increased layoffs.
Companies don't announce they're replacing departments with robots. They just don't hire the next three junior analysts. They don't backfill the retiring admin. They decide the next project needs two experienced developers working with agents instead of six working conventionally.
Five fewer positions here. Three unfilled vacancies there. An internship program that quietly shrinks.
The workforce can transform without a wave of pink slips.
The Stanford researchers were careful about their conclusions. Their findings are descriptive, they don't establish that AI caused the entire gap, and they found no evidence of economy-wide displacement attributable to AI. But the pattern still matters.
The first casualty of automation may not be the employee. It may be the job opening.
We're Automating the Thinking and the Clicking
Here's where the public conversation usually goes wrong.
Historically we traded physical effort for money. We built, moved, repaired, and manufactured things. Machinery made physical labor enormously more productive.
Then we built an immense category of work that's mostly cognitive. People get paid to understand information, make decisions, communicate, analyze, design, and coordinate. And it happens inside a computer. Read an email. Interpret a request. Find information. Make a decision. Open an app. Enter data. Produce a document. Communicate the result. A little thinking. A little typing. Some mousey-mouse work. Repeat until Friday.
I'm not dismissing the skill involved. I'm identifying the ingredients.
There's the cognitive component: understanding the problem, interpreting information, deciding what should happen. And the execution component: operating software to make it happen. Traditional automation was already good at the second when instructions were explicit and workflows were predictable. Agentic AI is increasingly capable of combining both. An agent can interpret an objective, gather information, plan, operate applications, write software, evaluate results, and adjust.
Is it perfect? Absolutely not. It makes mistakes, misunderstands context, and produces nonsense with impressive confidence. Reliable deployment still requires testing, verification, permissions, oversight, and someone accountable when it fails. But the economic threshold was never perfection.
If a system completes 70% of a workflow at substantially lower cost, that's meaningful. At 90%, it's transformative. The value of labor now depends less on how difficult a task is for a human and more on how difficult it is for an intelligent system.
That's a remarkable reversal. We spent generations tying cognitive complexity to economic value, and now some complicated cognitive work is easier to automate than simple physical work.
A sophisticated agent can analyze thousands of financial records and remain utterly incapable of fixing a leaking pipe. Your plumber may have a more defensible near-term labor market than someone who moves financial information between spreadsheets.
Welcome to 2026.
The Data Shows What Gets More Valuable
If the only story were elimination, this would be simple. Automated industries shrink, employment falls, workers get displaced. That's not consistently what's happening.
Revelio Labs' AI labor market tracker, updated October 28, found that companies identified as AI adopters saw approximately 27% greater headcount growth than non-adopters. But the growth was lopsided. Senior roles grew about 31%. Junior roles grew about 6%.
AI-adopting companies can grow while becoming significantly more selective about who they hire.
PwC's 2026 Global AI Jobs Barometer, analyzing more than a billion job advertisements across six continents, found that postings requiring AI skills carried an average wage premium of approximately 62% over otherwise comparable roles that didn't require them.
The same report found skills in highly AI-exposed occupations are changing more than twice as fast, and that entry-level postings in those fields were seven times more likely to require traditionally senior capabilities like judgment and leadership.
Let that sink in. Fewer doors. And the remaining doors expect you to already know things you used to learn inside.
The New Value Hierarchy
I think we're watching a reorganization of what makes human labor economically valuable. Not a perfectly defined hierarchy. A pattern.
Routine cognitive production is getting commoditized. Drafting standard communications, summarizing, extracting structured data, generating conventional code, formatting reports, predictable administrative work. Not worthless. Less scarce. And scarcity is where the premium lives.
So where does the premium go?
Domain expertise and contextual judgment. Knowing whether the output actually makes sense in a messy real world. Stanford's research supports this. Younger workers in occupations built on codified, standardized knowledge were especially exposed, while experienced workers leaning on tacit knowledge held up better. Not immunity. Different exposure.
Problem definition and orchestration. When execution gets cheap, deciding what to execute gets expensive. Someone has to spot the opportunity, understand the objective, find the constraints, connect the systems, weigh the tradeoffs, and decide whether the result is worth having. The economic unit shifts from the task to the outcome.
Trust and accountability. An agent can draft a contract. It can't accept the legal obligation of the person who signs it. It can prepare financial analysis. It can't carry fiduciary responsibility. It can generate a campaign. It can't promise it's truthful or protect a client's reputation. And it can't manufacture years of trust between two people.
Physical-world competence. Software is easy to automate because software can touch software. Pipes, panels, engines, and patients introduce constraints that don't care about a benchmark score.
Ownership of productive systems. This may be the big one. If intelligence becomes a service, who captures the value? The operator, the company that owns the workflow, the model provider, the data owner, the compute owner, the customer? These stopped being philosophical questions the moment they became contract terms.
In an increasingly automated economy, being productive and owning the means of productivity may produce dramatically different outcomes.
The Ratchet
Imagine two professionals selling the same service. One uses traditional tools and manual processes. One uses agents, integrated data, automated workflows, and a real verification process.
Both are competent. Both understand their industry. Both deliver good work. The second one does in a day what used to take several. Serves more customers, cuts costs, moves faster. Earns more, or works less, or builds a business around the leverage.
Wonderful.
Now the technology spreads.
Suddenly the extraordinary productivity is the baseline. Customers expect it. Competitors price against it. Employers staff for it.
AI proficiency produces an enormous individual advantage before becoming the minimum standard.
Which means early adopters capture disproportionate benefit, and adaptation itself never fully protects you. The market keeps moving. The next differentiation lands somewhere harder to copy: expertise, trust, proprietary capability, relationships, judgment, ownership.
The Apprenticeship Problem
Back to those entry-level jobs.
If the market increasingly rewards experience and judgment, we have a structural problem: how does anyone acquire them?
We learned by doing simple things before someone trusted us with complicated ones. Junior programmers fixed bugs before architecting systems. Junior accountants reconciled records before making judgments. Junior attorneys reviewed documents before leading litigation. Junior analysts built reports before directing investment decisions.
Some of that was tedious. Some could have been automated years ago.
It was also the apprenticeship.
If organizations eliminate it because they can buy the same output from software at a fraction of the cost, the immediate productivity gain looks wonderful. The bill arrives five or ten years later, when the organization needs experienced people who understand the underlying work.
We may be automating the system that produced the expertise we still say we value.
That's not an educational inconvenience. It's an economic bottleneck. A market that pays for judgment while offering fewer places to develop it isn't stable. We'll need new routes. Simulation. Project-based apprenticeships. Supervised agentic workflows. Real assessments. Faster paths from education into meaningful responsibility. The same technology that threatens the apprenticeship may become the best teaching tool we've ever had.
But that outcome isn't automatic. It has to be designed.
Productivity Up. Who Gets It?
The big question isn't whether AI raises productivity. It will.
It's who captures the gain.
If a company produces twice as much without doubling headcount, the benefit can flow to customers through lower prices, to workers through higher pay, to owners through profit, or to society through new activity.
Usually some mix, determined by competition, market structure, ownership, and policy.
Our economic arrangements assume wages. Most people get purchasing power by working or owning income-producing assets. So a world where productive capacity grows much faster than the human labor required to operate it isn't automatically a catastrophe. Cheaper output can create enormous consumer benefit and entirely new demand. But it's also not going to sort itself out because the technology is impressive.
The question isn't whether AI creates wealth. It's whether the mechanisms through which people participate in that wealth evolve alongside it.
What a Good Employee Means Now
The St. Louis Fed published something in September that reframes this. Between August 2024 and May 2026, the share of workers using generative AI for their jobs rose from about 33% to 45%.
Widespread. And frequently shallow. Plenty of people ask AI to summarize an email and call it adoption. Others run research, automate processes, build applications, and hand off significant chunks of their daily workload.
Same job title. Very different output.
Which is changing how employers think. Credentials and years of experience still matter. But the better question is becoming:
What can this person accomplish with the resources available to them?
Not how many hours. Not how many tasks they can grind through manually. What outcomes do they consistently deliver?
That's a shift from labor to leverage. And it's uncomfortable. Not everyone facing this has the same access, time, or security. An established professional with slack can experiment. Someone working two jobs can't. A company with millions for systems integration has advantages a small business doesn't. And nobody solves a restructured profession by watching a few tutorials. But the structural difficulty doesn't change the practical conclusion. Adaptation matters more, not less. We need better literacy, cheaper tools, real training, and more people who understand what these systems actually do.
Not everyone needs to build agents. But everyone increasingly needs to understand how automated intelligence touches their work. Because the market won't preserve an inefficient workflow out of respect for the twenty years somebody spent mastering it.
Data, Not Certainty
Let me be precise about what we have.
Evidence of changing hiring patterns. Evidence that AI adopters grow while shifting toward experienced roles. Evidence of large wage premiums. Measurements showing adoption spreading. Corporate and government plans treating automation as workforce strategy.
What we don't have is proof that we're entering irreversible mass technological unemployment. The research doesn't establish that, and we shouldn't pretend otherwise.
Some companies will shrink. Some will expand. Some professions will transform, some will contract, new categories will appear. Plenty of deployments will fail, and hiring a competent human will turn out cheaper than maintaining an elaborate system that doesn't work.
But uncertainty about the destination doesn't hide the direction of travel.
The significant development isn't that AI has eliminated the need for human labor. It's that the economics of producing useful work are changing, and we can finally see where the consequences show up first.
In hiring. In requirements. In organizational design. In capital. In what experience gets rewarded.
What I'm Watching
Not another countdown to the end of work. These instead.
Does hiring for routine cognitive work keep weakening?
Do organizations keep favoring people who can orchestrate intelligent systems?
Do AI wage premiums persist as the skills become commonplace, or become ordinary expectations?
Do small businesses get enough leverage to compete with incumbents?
Do productivity gains become lower prices, higher wages, more jobs, or mostly bigger profits?
And most importantly: do we build new ways to create expertise as the old ones get automated?
The Next Economy Won't Have Less Work. It'll Have Different Economics
I don't think we're heading toward a world where humans have nothing useful to contribute. There are too many unsolved problems and unmet needs for that. But I don't think the employment structures of the last century survive untouched either.
Routine cognitive production is under pressure. Expertise combined with automation is getting more valuable. Judgment, accountability, trust, and ownership are emerging as the differentiators. And the amount of useful work one person can do is decoupling from the number of hours it traditionally took.
Work, employment, productivity, and compensation may no longer maintain the relationships we got used to. For five years we asked whether AI would change the labor market. Turns out it already is. Not everywhere. Not uniformly. Not the way most people predicted.
The interesting question isn't whether machines can do work humans used to do. It's what happens to the value of that work once they can. We spent centuries building an economy around the scarcity of human labor and expertise. We're now building technology that makes enormous amounts of cognitive work cheap, repeatable, and available on demand.
The opportunity is extraordinary. So is the disruption. And we should stop discussing this as some distant possibility. The hiring patterns are already changing. The productivity gaps are already measurable. The price signals are already here.
We spent five years debating what would happen when intelligence became an industrial resource.
Now we're finding out.
Research and Reporting
Rich Washburn is a technologist, strategist, and Founder & Chief AI Architect of ARIA AI Labs, working at the intersection of AI, infrastructure, communications, and capital. He also serves as Managing Partner and Chief AI Officer at Eliakim Capital.
Comments