New jobs created by AI have raised the bar, and young people are losing the chance to gain experience in entry-level positions

[[**New professions created by AI keep emerging, but the threshold for ordinary workers to enter these new positions has not fallen—it has risen; more seriously, AI is replacing entry-level jobs, depriving young people of the chance to practice and accumulate experience, and the pipeline for cultivating the next generation of experts faces the risk of breaking.**

**Key points**

**1. New professions have high thresholds, making it hard for ordinary workers to transfer**

New positions such as digital twin and embodied intelligent robot application mostly require technical backgrounds in algorithms and programming; low-threshold positions such as data collectors offer limited pay and room for growth, leaving job transferees caught between insufficient skills and a drop in income.

**2. AI is replacing entry-level positions, and newcomers are losing the chance to practice**

Companies can now complete basic work that once required multiple people with "experienced employees + AI," and therefore hire fewer newcomers. Stanford research shows that in occupations highly exposed to AI, employment levels for young people aged 22-25 are about 19% lower than in occupations with low exposure.

**3. Experienced workers can transition using industry knowledge, while young people are trapped in the "experience paradox"**

**4. No one is paying the cost of cultivating the next generation of experts, and the industry faces a talent gap**

It is hard for companies to recoup training investment from newcomers' output, making them increasingly unwilling to cultivate talent; alternatives such as internships and simulated practice have limited effect, and if the responsibility for training falls on individuals, career opportunities will further tilt toward those with resources.2026-10-11 12:13The new jobs created by AI may never come your wayFang Leng Tu De LaQuick overview

This article is from the WeChat public account: Fang Leng Tu De La Shuzhai, author: Fang Leng Tu De La

A person who has worked in internet operations for several years recently wanted to change jobs.

Opening a job-hunting app, they found more and more AI-related positions: agent development, AI system application, robot training—each sounded promising. But clicking in, the well-paid ones required knowledge of algorithms, programming, and project experience; the lower-threshold ones might pay far less than the previous job.

Interestingly, while job seekers are still worrying about their next job, the list of new professions keeps getting longer. Recently, another batch of future-sounding professions entered public view.

On September 9, the Ministry of Human Resources and Social Security and other departments announced 11 new professions and 23 new job types, 5 of which are digital professions, including digital twin engineering technicians and embodied intelligent robot application technicians, while agent developers also entered the list of new job types. Profession names are increasingly tech-flavored, and ordinary people, before even figuring out what "digital twin" means, have begun worrying whether they will be replaced by it.

According to the experience of the past two hundred years, technological progress eliminates old jobs and also creates new ones. The steam engine brought factories, and computers gave rise to programmers and network engineers. But for an operations worker who just lost their job, someone who has done finance for over a decade, or a graduate looking for work, the addition of an "agent developer" to the occupation catalog does not mean they will get a new work badge tomorrow. After new jobs appear, how exactly do ordinary people from their original positions get in?

Over the past two hundred years, how have machines helped ordinary people find jobs?

1. Machines lowered the entry threshold: enter the factory first, then learn skills

In 18th-century Britain, textile work was a craft requiring long apprenticeship training, and production efficiency depended on the weaver's experience. After the spinning jenny and power loom appeared, production gradually concentrated in factories. Machines took over complex motions, and factories hired people to tend equipment and handle raw materials. Rural laborers with no textile experience might not understand the machine's structure, yet after training they could handle one of its processes.

The Second Industrial Revolution pushed this division of labor further. In 1913, Ford began using a moving assembly line, and automobile manufacturing was broken into standardized processes; assembly work that originally required skilled craftsmen was assigned to different stations. Workers only needed to be familiar with part of the operation to participate in making cars. A person who had just moved to the city might never have ridden in a car, yet could already work in an automobile factory. The assembly line fixed craft experience into the process and lowered the entry requirements for new workers.

As machines became more advanced, some positions demanded less and less personal craftsmanship. China after reform and opening up went through a similar process: manufacturing absorbed a large amount of rural labor lacking industrial experience through job training and standardized management. Industrialization provided the first career channel: machines lowered the entry threshold, and ordinary people could start work first and learn skills later.

2. Education plus entry-level positions: gradually becoming experts from newcomers

With electrification and the development of modern enterprises, work such as equipment maintenance and circuit design began to require systematic knowledge. Schooling therefore took on a more important role. The economic history research of Goldin and Katz found that the high school graduation rate of young Americans rose from about 9% in 1910 to 51% in 1940. More children from ordinary families first gained basic knowledge and then entered technical and managerial positions.

The computer revolution developed this growth model. After graduation, programmers start by writing simple code and fixing bugs, gradually understanding why systems fail and which changes may trigger chain problems. Finance staff grow from checking vouchers to judging business risks, and lawyers from searching cases to handling complex disputes, with similar growth paths.

These entry-level jobs carry little glamour. A programmer spends half a day finding a low-level error, auditors check numbers repeatedly, and at the time they feel like human tools; only years later do they realize that professional judgment grows slowly out of concrete problems. Newcomers touch real business, senior employees correct them, and the work itself also carries a training function. Newcomers create value for the company at the same time, which is why companies are willing to keep hiring.

The knowledge economy provides a second career pathway: enter a professional field through education, then accumulate experience through entry-level work and gradually become an expert. The industrial era allowed many people to work first and learn later; the knowledge economy has built a ladder from school to entry-level positions to senior positions. How many jobs new technologies can create is important, but whether ordinary people can find an entry point is equally important.

1. New occupations have increased, but the threshold for entering them has not necessarily fallen

Positions such as digital twin engineering, embodied intelligent robot applications, and AI system deployment often require technical knowledge or industry experience. Turning an administrative staff member with ten years of work experience into an algorithm engineer in a short period is probably no easier than getting an ordinary driver to switch careers and design engines. Occupational catalogs can be updated, but abilities cannot be upgraded with one click.

AI has also created jobs that are easy to enter. In September, a company in Henan recruited 80 robot data collectors and 20 reviewers on a university employment platform. The collectors wear motion-capture equipment, operate robotic arms to complete movements, and record data. The recruitment information showed that they could start a trial shift after two days of training, with a comprehensive monthly salary of 3,500–6,000 yuan.

In the past, masters trained apprentices; now humans act as masters for robots. But data collection and advanced AI development differ in income and room for growth. For knowledge workers who originally had stable incomes, switching to teaching robots to grab things may not be an ideal career upgrade.

A report released by the McKinsey Global Institute in September estimates that by 2035, about 11 million workers in the United States may need to change occupations, of whom about 14% have direct transition paths with relatively small skill gaps and no obvious pay cut. People familiar with traditional office processes must fill in knowledge and project experience if they want to enter AI technical positions; if they turn to low-threshold positions, they may have to accept a drop in income. AI has created new occupations, but it has not necessarily created the conditions for ordinary people to enter them.

2. As AI takes over basic work, newcomers may lose the chance to practice

Take programmers as an example. Experienced engineers can quickly locate faults and know how to make changes without affecting other systems; newcomers must read code line by line and debug repeatedly, and may spend half a day on a small error. Companies pay wages for this inefficiency, and they are also training future engineers.

AI coding assistants can generate code, locate errors, and write tests. Newcomers can complete tasks that used to take several hours just by stating requirements and checking results. But the process that originally helped them understand the problem—from checking logs and forming hypotheses to verifying causes—may be skipped entirely. The code can run, but they may not be able to explain clearly why it works. An experiment at the Technical University of Munich in 2026 involving 275 programming students found that students using AI performed better in exercises, but their knowledge mastery did not improve in tandem. A gap is beginning to appear between producing professional results with the help of AI and truly acquiring professional ability oneself.

Corporate demand for entry-level positions may also change accordingly. Research updated in August this year by the Stanford Digital Economy Lab, based on U.S. ADP payroll data, found that in occupations with high AI exposure, the employment level of employees aged 22–25 was about 19% lower than the growth trajectory of peers in low-exposure occupations, with the gap mainly reflected in reduced hiring of young employees.

This replacement can be very quiet. A company originally planned to hire five programmers, found that existing employees could complete new projects using AI, and therefore hired only two; when an operations employee leaves, the remaining people use AI to share the work, and the headcount is no longer replenished. There is no layoff announcement, and no one clears out a desk. In the past, when machines took your job, at least there might be a layoff notice; now AI leads companies to cancel hiring, and you do not even know that you originally had a chance to be hired.

3. Why is "experienced employees + AI" increasingly cost-effective for companies?

On October 8, Reuters reported that AT&T has handed document review, legal research, and other work previously given to junior lawyers at external law firms to an internal 38-person LegalEdge team working with AI. Billable hours for external lawyers have decreased accordingly, and AT&T hopes law firms will provide more strategy and professional judgment for complex cases.

For AT&T, basic work that previously required hourly fees can now be handled internally. But what law firms lose is not only billable hours. In the past, junior lawyers generated revenue through research and review, and law firms used it to pay wages and cover part of their training investment. Senior lawyers passed on experience to newcomers when reviewing results, and a few years later some of them became people who could handle cases independently. As clients buy fewer basic services, law firms also have fewer opportunities to train newcomers.

Other knowledge-based companies will do the same math: hiring newcomers requires paying wages and arranging mentoring, and only after several years may they be able to work independently, and the people trained may also job-hop. Experienced employees working with AI, by contrast, can complete tasks immediately. The problem of corporate training in economics has always existed: talent that one company pays to train may be hired away by another in the future. AI makes basic labor cheaper and also makes it harder to recover training investment through newcomers' output.

As a result, hiring requirements increasingly resemble a circular question: applying requires three years of experience, but the company that should have hired you three years ago also required three years of experience. Companies increasingly prefer experienced people, but may provide fewer and fewer positions that let people gain experience.

Today's senior programmers, auditors, and senior lawyers were all yesterday's newcomers. A single company hiring fewer people can improve current efficiency; but when the entire industry waits for others to train talent, how many young people capable of independently handling complex problems will remain in a few years? Training costs will not disappear; they will only become a problem that the entire industry faces together.

English text is being prepared.

1. Experienced people bring old skills into new occupations

Manufacturing companies use AI to predict equipment failures, but someone needs to understand the production line and the cost of downtime; retailers introducing intelligent replenishment need people who understand inventory and store sales; finance departments using AI to review expenses also need someone to judge which anomalies involve real business risks. Models can make suggestions, but whether those suggestions can be implemented still depends on people who understand the business to check them.

Manufacturing engineers can move into industrial AI applications, and finance staff can participate in the deployment of intelligent finance systems. Such transitions require adding tool capabilities, but do not require erasing more than a decade of work experience. A workshop supervisor who understands production scheduling, even without being able to develop models, can tell the technical team which suggestions cannot be implemented. When companies purchase AI systems, they ultimately need someone to connect them to the business and be responsible for the results of use. For mature workers, their original industry experience may be the most valuable ticket to entry into new AI jobs.

2. For young people, the first thing they need is an opportunity to gain experience

The problem is that young people who have just graduated do not have this ticket. In the past, companies allowed newcomers to start with simple tasks and learn while working; now they hope newcomers can use AI to independently take charge of projects as soon as they join, but may reduce the positions where people can practice. Young people have to prove they already have work ability before they can get their first job.

Some people gain exposure to real projects through school laboratories or internships, while others can only repeatedly practice simulations with AI. Tools are available to everyone, but opportunities for hands-on practice and professional guidance differ. In the past, people gained experience by entering companies; in the future, they may need experience before they can enter companies, and career competition has therefore moved earlier, to the level of schools and family resources.

3. Who will pay the cost of training the next generation of experts?

U.S. law firm Ropes & Gray this year allowed more than 1,000 lawyers to count up to 100 hours of AI experimentation and training toward annual billable requirements; previously it also arranged nearly 400 hours of learning and simulated practice for first-year lawyers. In the past, newcomers accumulated experience through basic work that could be billed to clients; now law firms are beginning to set aside time for learning itself.

Future training may require rotations on real projects and mentor guidance. AI can quickly provide answers, but identifying errors and assessing consequences still depend on practice. After AI completes basic tasks, companies need to specially arrange for newcomers to make judgments first, then use AI to verify, and someone still has to guide them. These arrangements all cost money, and whether companies, schools, or individuals pay will affect who can gain growth opportunities. If training increasingly depends on personal investment, people with resources will find it easier to cross the career threshold.

There are more and more new occupations; where will the next generation of experts come from?

Returning to the constantly updated list of new occupations at the beginning. In the past, new technologies created jobs and also provided ways for ordinary people to enter work: industrialization had factory jobs, and the computer era had school education and entry-level positions. Today, mature employees working with AI can take on more tasks, and companies save current labor costs, but cannot save society the time needed to train the next generation of experts.

An industry can temporarily hire fewer newcomers, but it cannot forever rely only on existing experts. Ten years from now, when professionals retire or change careers, where will that valuable experience come from? If the occupational list keeps growing but career entry points become narrower and narrower, technology will change not only the number of jobs, but also ordinary people's opportunities to gain a professional identity.

Has published 75 articles on Huxiu. This content is published with the author's authorization, and the views represent only the author and do not represent Huxiu's position. If you have objections or complaints about this manuscript, please contact tougao@huxiu.com. Everyone who is changing and wants to change the world is on the Huxiu APP[ ]