In the medium to long term, AI may (or may not) replace all human jobs. However, in the short term, AI does not seem to have achieved this yet. The employment rate for prime-age workers in the United States remains near an all-time high.
A recent survey of corporate CFOs found that there was “little evidence of short-term aggregate employment declines due to AI.” A survey of European companies has so far found no evidence of job cuts, despite productivity gains from AI. Jeffrey Hinton, one of the pioneers of modern AI, famously predicted that AI algorithms would lead to the imminent displacement of all radiologists. In fact, radiologists are in greater demand than ever before.
So even if AI might replace humans in the future, it is not happening now. But it is teeth The nature of your work may change. Software engineers, whose jobs were largely about writing code just a few months ago, are now primarily checkers and maintainers of code written by AI. However, this does not eliminate the need for software engineers. At least, not yet. Their job description has just changed.
Humlum and Vestergaard (2026) find that, at least in Denmark, this pattern of workers moving to new jobs without losing their jobs is the norm so far.
[M]most employers in [AI] At-risk occupations are adopting chatbot initiatives, workers are reporting increased productivity, and new AI-related tasks are becoming popular. Still…we estimate Exact zero effect on revenue and recorded time We exclude impacts greater than 2% two years after the start of ChatGPT at both the worker and workplace levels. What moves is the structure of work. Employers will absorb AI in the following ways: Reorganize tasksNew tasks such as content generation, AI monitoring, and AI integration will be included, and adopters will move into higher-paying occupations where AI chatbots are more relevant, but the numbers are still too small to beat the average income. [emphasis mine]
In other words, so far AI is replacing tasks, not jobs. Alex Imas and Soumitra Shukla write that we can expect this pattern to hold as long as there are some things that only humans can do. Observers of AI are constantly finding that its capabilities are “jagged,” meaning that it is much better at some tasks than others.
This is good news for people worried about losing their jobs (at least over the next 10 years). However, it remains a very vexing problem for those trying to decide what to study. Ten years ago, telling young people to “learn to code” made sense, or at least seemed to make sense. What would you tell them to learn now? What tasks will humans still need to perform, and which tasks will be encompassed by AI?
AI is steadily improving in such a wide range of tasks that it is difficult to predict what exactly. Even if people were certain they would do something, they would still be doing it five years later.
Some of my friends have spent the last decade or more thinking carefully about what the future of work will look like in the age of AI. No one has ever found a satisfactory answer. As AI technology develops and changes, even the most plausible predictions about the future of human work tend to be disproven almost as quickly as they are made.
But I’ve been thinking about this question for a long time, and I think I’ve begun to see an answer. I think that jobs in the near future will be broadly divided into three categories: office workers, specialists, and small and medium-sized enterprises.
Let’s talk about the experts first, because they are the easiest to understand. A new theory by Luis Garicano, Jin Li, and Yanhui Wu explains why some workers largely keep their current jobs.
Like many economists, Gallicano et al imagine a job as a bundle of different tasks. But they theorize that in some jobs, these tasks are only “weakly bundled” and the same person doesn’t actually need to do all of these tasks.
For these jobs, it’s easy to split tasks between multiple workers or between humans and AI. However, in other work, authors assume that tasks are “strongly bundled.” This means that the same person responsible for one part of the job must also do the other parts or the job cannot be completed.
The basic conclusion of this paper is that AI tends to replace weakly bundled jobs much faster than strongly bundled jobs. For example, we theorize that while most basic scan-reading tasks can be performed by AI, there is still work for radiologists because there is a lot of other work they need to do to provide patients with the care and expertise they seek.
They predict employment in highly constrained industries that resist automation until AI capabilities become very good.
People who are engaged in these highly constrained jobs are expert. An example of a specialist is a blogger. AI has so far gotten very good at doing background research, proofreading, and many other tasks that help with the writing process. However, although AI can generate an infinite amount of text, it is still not good at writing sentences.
Writing conveys a unique human perspective. Simply pressing a button and generating text doesn’t convey what you want to say. So the tasks that make up my own work, at least so far, are strongly bundled. AI is making me more productive, but I’m not at risk of losing my job right now.
But what happens to these weakly bundled jobs?Garricano et al. We predict that these will only start to decline until demand becomes sufficiently inelastic, that is, once AI productivity becomes so high that its output reaches a point of diminishing returns to consumers. After that, automation tends to replace human labor. Automation is no longer a way to make more things with the same amount of workers, but a way to make the same amount of things with fewer workers.
Until that point, growing demand will leave much more work to be done by people in less restrictive jobs. But at the same time, the “jagged” strengths and weaknesses of AI are constantly changing, so companies don’t know which tasks to hire employees for.
The rapidity with which Claude Code replaced the task of writing code illustrates this problem. By 2025, companies hiring software engineers will be able to judge their merits based on their ability to write code. In 2026, companies will need to judge the merits of software engineers based on how good they are at checking and maintaining code. These skills don’t always work together.
I think the solution is to hire more generalists. Rather than choosing people to perform specific tasks, companies will choose people whose job is to constantly learn what AI does and doesn’t do well and fill in the gaps. Cedric Savarese sums up this idea:
The first stage of “Vibe Freedom” is…[t]He was afraid that a report that would take him overnight would look better than anything you could do on your own in just a few minutes…The next stage comes almost out of the blue – there’s something not quite right. You begin to question the accuracy of your work. After reviewing it, I think it might have been faster to do it myself from the beginning…
You argue with the AI and are led down a confusing path, but slowly you begin to develop a deeper understanding of its mental model of mind. Learn to confidently recognize when you’re wrong, learn to refute and cross-check, learn to trust and verify…
Curiosity becomes essential. So is a willingness to learn quickly, think critically, spot contradictions, and rely on judgment rather than treating AI as infallible. That’s the new job of a generalist. It’s not about being an expert on everything, but understanding the AI’s mind enough to spot when something is wrong and follow the true specialists when the stakes are high.[.]
Fundamentally, AI becomes less reliable, but not in a predictable way. Its mistakes and shortcomings require continuous human investigation and correction. This is the job of a generalist. Companies will need to hire people who can do a little bit of everything instead of people who do “payroll,” “back-end engineering,” and “accounting” in case something goes wrong with AI.
In fact, there is an example in Japan of a corporate system that relies heavily on this type of generalist. Until recently, Japanese companies treated “salary workers” as a largely fungible workforce, requiring them to rotate between different departments and learn a wide range of tasks. You might start your career in human resources, then move to accounting, then product design, and so on.
This system may not have been very efficient, and the lack of professionalism may have contributed to Japan’s notoriously low white-collar productivity. And that may be the reason why the jobs of salaried workers have continued to decline for many years. But in the age of AI, it may finally make sense. As human expertise is replaced by AI expertise, humans may find themselves hopping from task to task, doing whatever the AI is bad at and overseeing whatever the AI is good at.
In other words, instead of hiring people like great accountants or great HR specialists, companies may start hiring people who are great AI wranglers and have the drive, mental flexibility, and energy level to keep plugging the ever-changing holes in AI capabilities. In other words, office worker.
Under the salaried employee system, the number of years of service is naturally long. If I’m a highly specialized engineer, I can take advantage of that talent and move to another company with my human capital intact.
But if I’m a generalist who does a little bit of everything, my internal network and understanding of the company’s systems are more important to my value as a worker. Because of this, my mobility is greatly reduced. I would like to stay at a company where my years of service are more valuable than new employees.
We can already see signs of this happening in corporate America. We are in a “don’t hire, don’t fire” economy. Workers are focused on their jobs and refuse to change jobs, and companies are keeping workers there instead of hiring new workers.
This is exactly what we would expect from a company-specific model of human capital, in other words, an economy where everyone increasingly recognizes that modern employees need to behave like Japanese office workers.
The hypothesis here is that people don’t want to quit their jobs (and companies are happy to keep them) because the rapid advances in AI may reduce the value of their technical skills. Instead, it’s still important for them to stay with the company and get to know the people and how things work.
In other words, America may begin to accept the salaryman path from now on. But the third category of future employment, self-employment and small businesses, is also very Japanese.
Japan has long had a very high ownership rate in small and medium-sized enterprises. It has one of the highest proportions of small and medium-sized enterprises in the world. In both manufacturing and retail industries, Japan has traditionally had far more small and medium-sized enterprises than other OECD countries. This proportion is currently on the decline as the population ages and business owners retire without successors or successors. But it could still point the way to an AI-enabled future.
Created by AI lever action;Small teams can do more. For many companies, the optimal size for this team is one or a few people. Therefore, I expect that many small businesses will emerge as people use AI agents to increase their productivity and require fewer (or no) employees.
In other words, I expect that AI will make America’s labor system look a little more like Japan’s labor system from the 1960s to the 2000s. There will be plenty of generalists running around looking for things to do in-house, there will be plenty of small-business businessmen who strike out on their own, and there will be a few specialists with specific skills that are still valuable.
If you’re not one of the lucky few in the latter category, your choices are either to become a cog in an ever-changing corporate organization, or to stand up on your own and manage an AI “team” to sell some goods or services directly to consumers.
This may not be the most optimistic or appealing view of the future of work, especially for people who have spent their lives thinking that their particular job skills are valuable to society. However, it may be better than humanity becoming economically obsolete.
This article was originally published on Noah Smith’s Noahpinion Substack and is kindly republished. Become a Noah Hopinion subscriber here.
