Full text 4,000 words | Reading approximately 13 minutes
Continuing to watch: Stanford Professor: AI Isn't About Wage Cuts, It's About Job Displacement, Young People Are Most Affected
(Erik Brynjolfsson of Stanford on the impact of AI on society)
Wages remain unchanged, but jobs are disappearing first.
A new research report released by Stanford's Digital Economy Lab on August 26 shows that the employment rate for young people aged 22–25 in professions heavily affected by AI has already dropped by 13%. It's not that wages have decreased, but that jobs have disappeared.
This is not a pessimistic prediction, but a conclusion directly derived from millions of payroll data. The research team used data from ADP, the largest payroll software provider in the United States, tracking the real employment status of millions of employees monthly, covering tens of thousands of companies, with data updated until July 2025.
Professions still exist, but the threshold has become higher.
——Erik Brynjolfsson, Stanford Professor, core insight from his research team
Why? AI isn't replacing entire professions, but rather the entry-level steps for new learners and those building experience.
The study also found:
- If AI is treated as an automation tool, entry-level positions will be compressed;
- If AI is treated as an augmenting assistant, experienced employees will drive hiring.
(Axios news headline, a Silicon Valley tech media outlet)
This finding has attracted attention from authoritative overseas media such as TIME and Axios, both pointing out that the first wave of AI's impact is falling on young people and junior positions.
ChatGPT was released just over two years ago, and the job market landscape has already been reshaped. This is a "payroll record" of the AI era, and a "career warning" for young people.
More importantly:
What does this mean for you? And how should you respond?
I. Four Facts: Who is Being Replaced by AI
Erik Brynjolfsson is one of the most frequently cited scholars in information economics, director of the Stanford Digital Economy Lab, and author of the bestseller "The Second Machine Age," a pioneer in researching the economic impact of AI and digital technology. He and his team in the report presented a seemingly calm, but in fact painful, data chart (shown below).
(A new study by Stanford University shows that AI causes more damage to certain types of jobs than others. Stanford Digital Economy Lab)
He used the most authentic first-hand data from the United States: 3.5 million+ people, monthly records, covering nearly 3 years. These are real records from payroll systems, not survey predictions.
The following four facts come from a direct analysis of this data and form the core of this report.
Fact One: Not layoffs, but hiring freezes
Brynjolfsson found:
The real crisis is not in layoffs, but in hiring freezes.
Many positions appear not to have had layoffs, but what's happening behind the scenes is that new hires are no longer being accepted.
From late 2022 to July this year, job opportunities for young people aged 22-25 in industries most affected by AI decreased by 13%. AI allows companies to accomplish more tasks with existing personnel, or to replace junior operations with tools. This change, without warning, is fatal to those just starting their careers.
For recent graduates, the challenge is not reduced salaries, but no entry point.
Fact Two: AI helps you, you stay; AI replaces you, you go
The research team divided jobs into two categories: those where AI helps, and those where AI replaces.
For example, a market analyst can quickly create reports using AI, which is enhancement; a junior content editor, AI can write directly, which is replacement.
Data shows that those who are "enhanced" have seen an increase in job opportunities and income; those who are "replaced" are seeing fewer entry opportunities.
The research team further stated:
What we're seeing is not AI impacting everything, but making some positions more advantageous and completely closing recruitment for others.
Fact Three: Impact transcends industry boundaries
Many people believe that AI's impact is limited to specific industries, such as media, translation, or customer service.
But Brynjolfsson said:
"The trend we're seeing isn't an industry-specific problem, but a common characteristic: entry-level tasks are simultaneously decreasing across many industries."
- Junior record keepers in the medical industry,
- Document organization assistants in the legal industry,
- Content annotators in the education industry…
These positions haven't disappeared, but they accommodate fewer and fewer people.
Why are young people bearing the brunt?
Because AI excels at tasks with standard answers and clear processes; and these tasks were precisely the first step for young people to enter the workplace and gain experience.
Now, this step is gone.
Fact Four: Traditional data indicators have blind spots
Traditionally, when we look at whether a technology affects employment, we often look at two indicators: has GDP grown? Are there still many job postings on recruitment websites?
But in the face of AI, both of these indicators are starting to become sluggish.
Professor Brynjolfsson pointed out: AI's impact is gradual; it looks like there's no problem, but in reality, changes have already begun.
- Positions are still advertised, but no one is actually hired;
- Wages appear stable, but tasks have shifted from new hires to AI or experienced employees.
Traditional data can't see this job loss, but payrolls don't lie.
These four facts tell us:
AI brings not a sudden collapse, but a quiet sinking of the foundation, and the first to lose their footing are young people just starting their careers.
The question is, with such a big change, why haven't many people noticed?
II. The Entry Point is Changing, Many Haven't Noticed Yet
Many people haven't realized the changes brought by AI because it lacks the traditional upheaval of technological replacement:
No layoffs were announced, no wage increases, many companies are quietly stopping recruitment.
Professor Brynjolfsson mentioned in an interview:
"What we're seeing isn't everyone being replaced by machines, but companies are using AI to have existing senior employees complete tasks that originally belonged to young people, alongside various AI tools."
If you only look at GDP data or browse recruitment websites, you might think everything is normal.
This study used real payroll records and uncovered a hidden problem:
"AI reallocates tasks that originally belonged to new hires to machines and experienced employees."
They observed a real example:
- Previously, a startup would hire two interns to organize market data;
- Now, an AI tool completes the initial screening in 5 minutes, then hands it over to the project manager;
- The intern position is still listed on recruitment websites, but no one is hired long-term because there's no longer anyone to assign tasks for this role.
This change appears calm, but it is crucial.
The professor concluded:
"What AI is changing is not the total volume of the labor market, but who can enter, from where they enter, and whether they can advance after entering."
You don't see it, so you underestimate it
GDP fundamentally measures valuable transactions, meaning how much money you've spent and how much companies have earned.
But Brynjolfsson pointed out:
AI brings a large amount of "invisible value," especially zero-priced goods.
For example:
- You don't pay for Google Maps, but you can't live without it;
- You don't pay for ChatGPT, but it has already helped you write countless proposals;
- Companies use AI to process junior content, eliminating the need to hire another person, but GDP won't record the action of "not hiring this person."
In Professor Brynjolfsson's view:
"AI brings a lot of invisible value, especially those services you don't pay for but which change the way you work."
This explains a misunderstanding:
Just because economic data looks stable doesn't mean the impact doesn't exist.
In fact, the impact is quietly happening, but traditional indicators can't capture it. And this "invisible change" primarily affects young people who haven't yet built networks, accumulated resources, or gained experience.
It's precisely for this reason that many people still think "the AI impact hasn't arrived yet."
III. Miscalculation Leads to Overlooking the Impact
Many people still feel that "AI's impact isn't that big" simply because they are looking for evidence in the wrong places.
While everyone is still using "are there layoffs?" and "are there fewer jobs?" to measure AI's impact, Professor Brynjolfsson's team is already using a new approach to examine this issue.
He said:
Many AI services are free. But not charging doesn't mean no value.
In the past, when we calculated the economy using GDP, we always used "price" as the measuring standard.
But in the AI era, value often occurs where you haven't paid.
So he proposed a new approach called GDP-B (B stands for Benefits), meaning: in addition to the income in GDP, we must also add "the benefits people gain from goods or services."
How to measure it? Not by subjective imagination, but through real "willingness to compensate" surveys.
Simply put: how much money would we have to pay you for you to be willing to give up a certain product or service?
The professor gave a classic example:
"We asked users: how much compensation would you need to go a month without air conditioning? The result showed that half of the people would be satisfied with $34, but some insisted on $2,000."
They used the same method to measure the true perceived value of many seemingly "free" things:
There's also a particularly typical saying:
Many people can go a month without electricity, but without electricity and without a mobile phone, they directly collapse.
Therefore, GDP-B is not a figment of imagination, but an estimation of its actual value from what users themselves say, "I can't live without this."
The true cost, both visible and invisible
The professor also specifically mentioned:
"Netflix's subscription fee might only be $15, but we measured that people's 'psychological compensation' they'd be willing to accept to go a month without it is $166."
Behind this is the concept of "consumer surplus": you spend less, enjoy more, and that extra part is the real sense of happiness, which traditional GDP completely overlooks.
In the AI era, this part has been greatly expanded:
- ChatGPT, Gemini, Claude, these AI-related tools are free, or only cost a few dollars
- But the help they provide, the time saved, and the labor they replace already exceed their price.
At this point, Brynjolfsson said something very interesting:
"Many people now secretly use AI to help them write answers when filling out questionnaires. If that's the case, why don't we just let AI be the user itself?"
So they designed a new approach:
- Turning AI into a "person": telling it, "You are a 25-year-old Black man living in Boston," or "You are a university professor living in New York."
- Then letting AI "answer" questions about consumer preferences, choices, and what services it would be willing to give up.
They call this: Large Language Model Agent (LLM Agent).
Even more astonishing, they later had AI "travel through time": pretending to be users from 2016 or 2017 to "evaluate how much Facebook was worth then."
The comparison chart shown by the professor illustrates:
The answers simulated by AI were almost identical to the questionnaires filled out by real users in 2016.
What insights did this bring?
He summarized:
"We are not using AI to replace people, but to expand into areas we previously couldn't see."
This agent model has already shown great potential, for example:
- To study a remote population, AI can simulate it in seconds, eliminating the need for months of on-site research;
- To predict how someone who has never used ChatGPT would evaluate it, AI can help you "imagine";
- It can even simulate consumer behavior across different countries and educational backgrounds.
This not only greatly improves efficiency but also allows "invisible value" to be calculated for the first time.
Based on these findings, Brynjolfsson's team put forward a profound reflection:
"In the process of AI reshaping the economy, are our measurement tools overlooking the changes that most need to be seen?"
IV. Jobs Disappeared, It's Not Because You're Not Good Enough
(Stanford Professor and co-author of the research report, Erik Brynjolfsson)
And these people are precisely those young individuals who are blocked from entering before they even start.
At the end of the interview, Professor Brynjolfsson returned to people, especially young people. He said:
AI isn't taking anyone's job; it's preventing newcomers from even getting a foot in the door.
This statement hits home precisely. Many young people feel they are not strong enough, not fast enough, not smart enough, but the problem is not there at all.
So it's not "you win and you stay," but "you don't get a chance to be seen."
AI hasn't taken away your abilities, but it has indeed changed everyone's career starting line.
The professor pointed out that the impact of AI is comparable to the most significant technological revolution in a century, but the most crucial aspect is that it changes everyone's starting point for entering the workforce.
Previously, skills could be gradually developed on the job.
Now, the job threshold has become higher; you need to be able to deliver results from the start, preferably also bringing content, traffic, and judgment.
If you are a recent graduate, changing careers, or just starting a new job, and you lack experience and don't know how to use AI tools, not finding a job doesn't mean you have a problem; it means the entire employment ecosystem has changed.
No "One-Size-Fits-All Advice," but He Said Three Things Particularly Worth Hearing
The first thing: You need to know which type of job you're in
He said they categorize jobs into two types:
One type is where AI can help, such as market planning or project research. People in these roles will become more valuable because they know how to use AI;
The other type is where AI can directly replace, such as initial drafting of documents, data cleaning, or repetitive operations. These positions themselves are no longer growing.
Which type is the job you want to do?
If you haven't started looking for a job yet, this question must be clarified first.
The second thing: What's truly valuable is what AI cannot imitate
The professor gave an example:
"AI can write reports quickly, but it cannot replicate the feeling of interaction you have when communicating face-to-face with a client; it can research information, but it doesn't know what direction the boss truly wants for a solution."
These "unspoken things," what the professor calls "tacit knowledge," are the hardest for AI to approach.
And this knowledge can only be learned through experience.
If AI takes away the entry-level positions where you would learn these experiences, then you need to actively seek out jobs that require interpersonal communication and flexible problem-solving.
It doesn't necessarily have to be high-paying, but it must involve dealing with people and offer opportunities to practice in complex situations.
The third thing: You need to master not AI technology, but the habit of using AI to do things
The professor provided clear guidance:
"It's not about everyone training models, but about people understanding when to use AI and how to apply it in their own work."
Just like you don't need to be able to build a car to drive one, but you must know when to press the accelerator and when to hit the brakes.
This is also the foundation of an "AI-augmented talent": not being good at technology, but knowing how to use it, daring to use it, and seamlessly integrating it into one's work style.
If you are a recent graduate, changing careers, or just starting a new job, what you need to do now is to find entry points that are still open to newcomers.
Remember, this job-seeking difficulty is not your problem, but a change in the entire workplace ecosystem.
Conclusion: AI Rewrites the Starting Line
The true impact of AI is not that everyone will be "unemployed," but that some people will never be able to "enter" the workforce.
This is not a future warning; it is a current reality. Stanford's data has clearly told us: the change has begun.
The most cruel part is that this elimination comes silently.
- No layoff notices,
- No salary cut announcements,
- Just those positions originally reserved for new hires, quietly disappearing.
This transformation first impacts young people, but it won't stop there. Next, more levels, more industries, and more ordinary people will face the same challenges.
If AI truly is the new industrial revolution, then we urgently need to redesign the "entry mechanism."
Otherwise, those who haven't even started running won't even find the starting line.
📮 This content is compiled from the research report "Canaries in the Coal Mine: Early Evidence for the Effects of Generative AI on the U.S. Labor Market" and related academic interviews published by Professor Erik Brynjolfsson's team at Stanford University's Digital Economy Lab. Unauthorized reproduction is prohibited.
Reference Materials:
https://www.youtube.com/watch?v=m59xi83O-gs&t=419s&ab_channel=EconomicStatisticsCentreofExcellence
https://www.youtube.com/watch?v=SstOWPTg7hk&t=12s&ab_channel=EconomicStatisticsCentreofExcellence
https://digitaleconomy.stanford.edu/publications/canaries-in-the-coal-mine/
https://digitaleconomy.stanford.edu/wp-content/uploads/2025/08/Canaries_BrynjolfssonChandarChen.pdf
https://time.com/7312205/ai-jobs-stanford/?utm_source=chatgpt.com
https://www.axios.com/2025/08/26/ai-entry-level-jobs?utm_source=chatgpt.com
Source: Official Media/Online News
Layout: Atlas
Editor: Shen Si
Chief Editor: Turing