For most of the industrial age, labor was relatively easy to count.
A company employed 500 people. A warehouse ran three shifts. A consulting firm billed 10,000 professional hours. Productivity could be expressed as output per worker, revenue per employee, or units produced per hour.
Artificial intelligence is beginning to break that arithmetic.
We are entering the age of Digital Labor: software systems capable of performing meaningful portions of work that previously required human attention, judgment, communication, analysis, or execution.
Not just software that stores information.
Not simply automation that follows an “if this, then that” rule.
Digital labor can increasingly research, write, code, analyze, organize, communicate, monitor systems, review documents, prepare decisions, interact with other software and, through AI agents, execute sequences of work with decreasing human involvement.
That distinction matters because the conversation about AI and employment is usually framed incorrectly.
The question is no longer simply:
How many jobs will AI replace?
The better question may be:
What percentage of the world's work will increasingly be performed by something that isn't human?
Those are very different questions.
The Numbers Are Already Significant
The International Labour Organization estimated in 2025 that approximately 25% of global employment is in occupations with some degree of exposure to generative AI. In high-income countries, that rises to approximately 34%, compared with around 11% in low-income economies. Only about 3.3% of global employment, however, falls into the ILO's highest exposure category.
That distinction is important.
Twenty-five percent exposure does not mean 25% unemployment.
Exposure means that portions of the work can potentially be performed, accelerated, modified, or automated by AI.
The IMF uses a broader methodology and estimates that nearly 40% of global employment is exposed to AI, rising to approximately 60% in advanced economies. The IMF estimates that roughly half of those exposed jobs may benefit from AI complementing workers, while the other half could face reduced labor demand as machines take over meaningful portions of their responsibilities.
Again, exposure is not displacement.
But neither is it insignificant.
Consider what 40% exposure means economically.
It means that a technological system capable of producing cognitive labor is touching a portion of work representing nearly half of the global employment structure.
That is extraordinary.
And we are still near the beginning.
Digital Labor Is Already Producing Real Productivity
The positive argument for digital labor is powerful because it is increasingly supported by evidence.
A major study of 5,179 customer-support workers found that access to a generative-AI assistant increased productivity by approximately 14% on average.
But the average hides something much more interesting.
Productivity among novice and lower-skilled workers increased approximately 34%.
Experienced, high-performing workers gained comparatively little.
AI, in other words, appeared to distribute some of the accumulated knowledge of the organization's best workers throughout the workforce.
That is potentially revolutionary.
Microsoft researchers studying thousands of knowledge workers similarly found measurable time savings. Workers with access to generative AI spent less time processing email and completed documents approximately 12% faster. Another randomized study found active users spent roughly three fewer hours per week on email, although effects on collaborative activities such as meetings were much smaller.
Three hours does not sound revolutionary.
Multiply it across 100 million knowledge workers.
That becomes 300 million hours of human attention potentially released every week.
That is the economic promise of digital labor.
It doesn't necessarily remove the human.
It removes portions of the work surrounding the human.
Then Comes the Uncomfortable Part
The first measurable labor-market effects are beginning to appear, and they are not evenly distributed.
Stanford's Digital Economy Lab analyzed payroll records covering millions of American workers and found no evidence, as of mid-2026, of widespread economy-wide employment collapse attributable to AI.
That should temper the most dramatic predictions.
But researchers found something else.
Among workers aged 22–25 in highly AI-exposed occupations, employment stood approximately 19% below where it would have been had it kept pace with employment among similarly aged workers in less AI-exposed occupations.
The researchers found that the difference appeared to be driven primarily by reduced hiring, rather than mass layoffs. Experienced workers showed no comparable decline.
That may be one of the most important early signals of the digital-labor economy.
The first casualty may not be the existing employee.
It may be the employee who was never hired.
Organizations historically needed junior analysts, junior programmers, assistants and apprentices because senior employees couldn't economically perform every intermediate task themselves.
Digital labor changes that equation.
One capable employee with five digital workers may eventually accomplish what once required a department.
That doesn't necessarily mean five people are fired.
It might mean four positions are simply never created.
Those missing jobs are much harder to see.
The Apprenticeship Problem
This introduces an even deeper question.
Today's senior professionals became senior professionals because yesterday they were juniors.
They reviewed documents.
They reconciled spreadsheets.
They wrote basic code.
They prepared presentations.
They answered repetitive questions.
They made mistakes under supervision.
Those seemingly inefficient activities created experience.
If digital labor absorbs the bottom 30% of an occupation's workload, businesses may gain tremendous productivity today while unintentionally eliminating the training ground that produces tomorrow's experts.
That problem has no historical dataset because we have never experienced it before at this scale.
We do not know what happens when millions of people outsource portions of thinking, writing, memory, research and problem solving to machines beginning in school and continuing throughout their careers.
We don't know what happens to institutional expertise.
We don't know what happens to judgment.
We don't know whether humans become dramatically more capable because AI amplifies them—or progressively less capable because AI performs the repetitions through which capability was once acquired.
Both outcomes are plausible.
The Percentage Narrative Can Mislead Us
The World Economic Forum estimates that broad labor-market transformation through 2030 could create approximately 170 million jobs while displacing 92 million, producing a net increase of about 78 million jobs.
That sounds reassuring.
But those changes represent approximately 22% of today's formal employment, and employers simultaneously expect roughly 39% of workers' existing skills to change or become outdated by 2030.
A net-positive employment number therefore doesn't mean an easy transition.
Imagine an economy eliminating eight jobs while creating fourteen completely different jobs.
Statistically, employment increased by six.
For the eight displaced workers who don't possess the qualifications for the fourteen new positions, the macroeconomic success story may feel considerably different.
Economies experience percentages.
People experience disruption.
Digital Labor Will Probably Create More Work Too
There is another side to this transformation.
History repeatedly shows that increasing productivity changes demand.
Cheap computing didn't eliminate computing jobs. It created industries that were previously impossible.
Digital labor could similarly produce enormous categories of activity that would be economically irrational using human labor alone.
A five-person company may soon perform market research comparable to a corporation.
A researcher may run hundreds of analyses simultaneously.
An entrepreneur may operate sales, support, accounting, marketing and software-development agents continuously.
Doctors, engineers, scientists, teachers and tradespeople may gain systems capable of handling large amounts of administrative overhead.
Entire businesses may become viable because the marginal cost of certain forms of intelligence approaches the marginal cost of computing.
That could produce an explosion of entrepreneurship and individual leverage.
The optimistic case for digital labor isn't simply fewer employees.
It is more capability per human being.
The Unknown Is Larger Than Either Side Admits
This is where certainty becomes dangerous.
Anyone confidently predicting massive permanent unemployment is extrapolating beyond the evidence.
Anyone confidently predicting that AI will simply create better jobs for everyone is doing exactly the same thing.
We don't know.
Anthropic's own analysis of AI usage illustrates the ambiguity. In recent data, approximately 52% of Claude interactions were categorized as augmentation and 45% as automation. Earlier measurements varied, suggesting the relationship between humans and AI is still evolving rapidly.
That ratio may ultimately matter more than any model benchmark.
If digital labor primarily augments humans, we may experience one of history's greatest productivity expansions.
If it increasingly substitutes for humans, the economic consequences become considerably more complicated.
Most likely, both happen simultaneously.
Different occupations.
Different companies.
Different countries.
Different workers.
Different outcomes.
Labor Is Becoming Something We Can Manufacture
Industrialization gave humanity manufactured physical power.
Software gave us manufactured calculation.
AI may give us manufactured cognitive labor.
That is a fundamentally different economic resource.
A company may soon have 100 human employees and 10,000 persistent AI processes conducting research, monitoring operations, generating software, analyzing customers and preparing decisions continuously.
How many workers does that company employ?
One hundred?
Ten thousand one hundred?
The vocabulary of economics hasn't caught up yet.
But the underlying transition has already begun.
Digital labor is not arriving someday.
It is entering organizations one task at a time.
The challenge is not simply protecting human jobs from machines.
It is determining how humans remain economically valuable, intellectually capable and ultimately responsible in a world where intelligence itself is becoming something we can deploy like infrastructure.
That may turn out to be one of the defining economic questions of the AI era.