Artificial general intelligence is usually described as AI capable of performing most economically valuable cognitive work at or above the level of humans. OpenAI, for example, defines AGI as highly autonomous systems that outperform humans at most economically valuable work.
Nobody knows when—or whether—we will reach that point. But I think the economic implications are worth taking seriously before the technology gets there.
The question that interests me most is simple:
What happens to an economy built around human labor if intelligence stops being scarce?
For most of history, useful human intelligence has been expensive. A company that wants more analysis, engineering, accounting, design, legal work, management, or customer support has traditionally needed more people. Software changed that relationship somewhat, but people still operated the software.
AGI could change the equation much more fundamentally.
If a capable AI system can perform an hour of knowledge work for pennies, operate continuously, copy itself almost instantly, and improve alongside the underlying models, then companies may no longer need to increase headcount in proportion to the amount of cognitive work they want done.
That would be a very different economy.
We are already seeing the beginning
Current AI is not AGI, and it is important not to confuse the two.
Most evidence today points toward task transformation rather than wholesale job replacement. The International Labour Organization estimates that roughly one in four jobs globally has some exposure to generative AI, while emphasizing that exposure does not mean a job will disappear. Transformation is currently considered more likely than complete replacement.
The IMF reaches a similar but broader conclusion. It estimates that around 40% of global employment is exposed to AI, rising to roughly 60% in advanced economies. Some workers could become significantly more productive. Others could see AI perform tasks that previously justified part of their employment.
And there are already measurable productivity improvements.
Stanford’s 2026 AI Index summarizes research finding productivity gains of around 14–15% in customer support, 26% in software development, and considerably larger increases in some marketing tasks. At the same time, the effects aren’t uniform: AI performs much better when work is structured, measurable, and provides clear feedback.
That distinction matters.
Today’s AI is often a tool used by a worker.
AGI, if it arrives in the stronger sense of the word, could increasingly become the worker.
The first impact may be fewer entry-level jobs
I don’t think the labor market necessarily changes because millions of people suddenly receive termination notices one morning.
There is a quieter way this can happen.
Companies simply stop hiring as many people.
If five analysts using AI can produce what once required ten analysts, a company doesn’t necessarily fire half the department immediately. It may hire fewer analysts the following year. Attrition takes care of some of the rest.
That could make entry-level jobs particularly vulnerable.
Stanford’s 2026 AI Index already points to labor-market effects being concentrated partly among younger workers in AI-exposed occupations. It reports that employment among software developers aged 22–25 has fallen substantially since 2024, while one-third of organizations surveyed expect AI to reduce their workforce over the coming year. Large-scale job losses have not yet appeared across the overall labor market, but hiring pipelines may be an earlier place to watch.
This raises a problem that I don’t think gets enough attention.
Entry-level work isn’t just cheap labor. It is how people become experienced.
Junior analysts become senior analysts. Junior developers become architects. Associates become executives.
If AI automates much of the work traditionally performed at the bottom of the ladder, companies could eventually face a strange problem: fewer places for humans to learn the skills required at the top.
Productivity could explode without wages doing the same
The optimistic AGI scenario is easy to imagine.
Companies become dramatically more productive. Scientific research accelerates. Software becomes cheaper to build. Administrative costs fall. New companies can operate with tiny teams. Goods and services become cheaper. Economic output rises.
There is a real possibility of enormous abundance.
But increased productivity doesn’t automatically mean that every worker becomes richer.
Suppose a company previously needed 1,000 employees and can eventually produce more output with 200 employees plus AI systems.
The company may become far more productive.
The remaining 200 workers might earn more.
Customers might pay lower prices.
Shareholders might make considerably more money.
But the other 800 workers still need somewhere to go.
The economic question therefore isn’t only:
How productive will AGI make us?
It’s also:
Who owns the systems producing that productivity?
If AI behaves primarily as a complement to human labor, workers can capture some of the gains through higher productivity and wages.
If AI behaves increasingly as a substitute for human labor, a larger share of economic value may flow toward whoever owns the models, compute, energy, intellectual property, data, and businesses deploying them.
That could make ownership of capital considerably more important relative to ownership of labor.
Not every job disappears
There is a tendency to jump from “AGI can perform knowledge work” to “nobody will have a job.”
I don’t think the transition would be that clean.
The physical world moves more slowly than software.
Electricians, construction workers, nurses, mechanics, technicians, plumbers, and countless other jobs require manipulating unpredictable physical environments. Robotics will continue improving, but deploying millions of reliable machines throughout the real economy is different from deploying another software agent in a data center.
Even in knowledge work, there are reasons humans may remain involved.
Responsibility matters.
Trust matters.
Regulation matters.
People may simply prefer dealing with other people in certain situations.
A bank might be capable of having an AI approve a complicated loan while still requiring a human officer to accept responsibility for the decision. A patient may want a human doctor involved even if an AI is better at diagnosis. A CEO might rely heavily on AI analysis while still being legally and socially responsible for the final decision.
The boundary between what AI can do and what society allows AI to do may become increasingly important.
There is also a much less dramatic possibility
All of this depends on AGI actually being as economically transformative as its strongest proponents expect.
That isn’t guaranteed.
Economist Daron Acemoglu has argued that the productivity effects of today’s AI could be meaningful but much smaller than the most aggressive predictions. His estimates suggest relatively modest total-factor-productivity gains over a ten-year period if AI mostly automates a limited set of tasks rather than producing fundamentally new capabilities.
That is an important counterweight to the hype.
We could discover that many jobs contain more tacit knowledge, organizational context, interpersonal judgment, and messy real-world decision-making than benchmarks suggest.
AI might become extraordinary without making humans economically obsolete.
Current usage also still looks heavily collaborative. Anthropic’s Economic Index found that slightly more than half of sampled Claude.ai interactions involved augmentation—people learning, iterating, or working with the model—rather than simply handing the entire task over. Automation was more common in API usage, where AI is deliberately embedded into software processes.
The future could therefore look less like “AI replaces everyone” and more like organizations being rebuilt around much smaller groups of highly leveraged people.
The question I keep coming back to
I used to think the main question about AI and employment was:
Which jobs will disappear?
I increasingly think that’s too narrow.
A more useful question might be:
What happens to the value of human labor when intelligence can be reproduced at near-zero marginal cost?
Even if most people remain employed, the structure of careers could change.
Companies may become smaller.
Individual workers may become dramatically more capable.
Starting businesses could become easier.
Some professions may shrink while entirely new ones emerge.
The advantage of being technically skilled may shift from knowing how to perform every task yourself toward knowing what to build, what questions to ask, what decisions matter, and how to direct increasingly capable systems.
And if AGI eventually does outperform humans across most economically valuable cognitive work, the challenge becomes larger than career planning.
At that point, we would have to reconsider how income is distributed in an economy where human labor is no longer the primary constraint on production.
Maybe that produces extraordinary abundance.
Maybe it produces extraordinary inequality.
Most likely, it produces some uncomfortable combination of both before society figures out the rules.
We’re not there yet.
But given how quickly the capabilities are improving, I think it’s time to start thinking about what happens when we get there.