Which workers will AI hurt the most? Young or experienced workers?

When Amazon CEO Andy Jassy wrote last month that he expected the company's use of artificial intelligence to "reduce our total workforce" in the coming years, he confirmed many workers' fears that AI would replace them. That fear was reinforced two weeks later when Microsoft announced it would lay off about 9,000 people, roughly 4 percent of its workforce.

That AI is about to displace white-collar workers is indisputable. But what kind of workers exactly? Jassy's announcement came amid a debate on precisely this question.

YOU MIGHT BE INTERESTED IN: Do your children talk to AI? These are the 10 key points you should know.

Some experts argue that AI is most likely to affect entry-level workers, whose tasks are often the simplest and therefore easiest to automate. Dario Amodei, CEO of AI firm Anthropic, recently told Axios that the technology could cannibalize half of all entry-level jobs within five years. Rising unemployment among recent graduates has compounded this concern, though this doesn't prove AI is the cause of their labor market difficulties.

But other AI industry leaders have taken the opposite view, arguing that younger workers are likely to benefit from AI and that experienced workers will ultimately be the most vulnerable. In an interview at a New York Times event in late June, Brad Lightcap, chief operating officer of OpenAI, suggested that the technology could pose problems for “a class of workers who I think are more tenured, more oriented toward a routine in a certain way of doing things.”

The definitive answer to this question will have far-reaching implications. If entry-level jobs are the most at risk, it may require a rethinking of how we educate college students, or even the value of college itself. And if older workers are the most at risk, it could lead to economic and even political instability as large-scale layoffs become a persistent feature of the labor market.

David Furlonger, a vice president at research firm Gartner who helps oversee its CEO survey, has considered the implications if AI displaces more experienced workers.

“What are these people going to do? How will they finance themselves? What is the impact on tax revenue?” he said. “I imagine governments are thinking about that.”

Is it true that AI is making managers better?

Economists and other experts who study AI often draw different conclusions about who it is most likely to displace.

If we focus on the fields where AI has been most widely deployed so far, the picture tends to be bleak for entry-level workers. Data from ADP, the payroll processing company, shows that in computer-related fields, employment of workers with less than two years' experience peaked in 2023 and has declined by 20 to 25 percent since then. The pattern is similar among customer service representatives, who are also increasingly reliant on AI.

Over the same period, employment in these sectors has increased for workers with two or more years of seniority, according to Ruyu Chen, a researcher at Stanford University who analyzed the data.

Other studies point in a similar direction, albeit indirectly. In early 2023, Italy temporarily banned ChatGPT, which software developers there used to help them code. A team of researchers from the University of California, Irvine, and Chapman University compared the change in Italian programmers' productivity with the productivity of programmers in France and Portugal, which did not ban the software, to isolate ChatGPT's impact.

While the study didn't analyze job losses, it did find that the AI ​​tool had transformed mid-level workers' jobs more favorably than those of entry-level workers. According to the researchers, novice coders used AI to complete their tasks slightly more quickly; experienced coders tended to use it to benefit their teams more broadly. For example, AI helped mid-level coders review other coders' work and suggest improvements, and contribute to projects in languages ​​they didn't know.

“When people are really good at something, they end up helping others rather than working on their own projects,” said Sarah Bana, one of the paper’s authors, adding that AI essentially reinforced this tendency. Bana said the paper’s findings suggested that AI would drive companies to hire fewer entry-level programmers (because fewer would be needed to complete entry-level tasks) but more mid-level programmers (because AI amplified their value for the entire team).

On the other hand, Danielle Li, an economist at the Massachusetts Institute of Technology who studies the use of AI in the workplace, says there are situations where AI could harm more skilled workers than beginners. The reason is that it can effectively decouple valuable skills from the humans who have traditionally possessed them. For example, it may no longer be necessary to be an engineer to program, or a lawyer to write a legal brief.

“That state of the world isn't good for experienced workers,” he said. “You're being paid for the rarity of your skill, and what's happening is that AI allows that skill to live externally to people.”

Li said AI wouldn't necessarily be good for less experienced workers either. But he speculated that the spike in unemployment among recent college graduates was due to employers' expectations that they will need fewer workers overall in the AI ​​era, not just fewer entry-level workers. A general hiring slowdown could have a bigger impact on workers fresh out of college, since they don't have jobs to begin with.

Robert Plotkin, a partner at a small intellectual property law firm, said AI hadn't affected his firm's need for lower-skilled workers, such as paralegals, who format documents his firm submits to the patent office. But his firm now uses about half as many contract attorneys, including some with several years of experience, as it did a few years ago, before generative AI became available, he added.

These more experienced attorneys draft patent applications for clients, which Plotkin reviews and asks them to modify. But he can often draft applications more effectively with the help of an AI assistant, except when the patent relates to a field of science or technology with which he is unfamiliar.

“I’ve become very efficient at using AI as a tool to help me draft applications, which has reduced our need to hire lawyers,” Plotkin said.

Some of the companies at the forefront of AI adoption appear to have made similar calculations, laying off experienced employees rather than simply hiring fewer entry-level workers. Google, Meta, and Amazon have all made layoffs since 2022. Two months before its most recent layoff announcement, Microsoft had laid off 6,000 employees, many of them software developers, while the July layoffs included many middle managers.

“Anything that's administrative, spreadsheet-related, email-related, document-management-type, AI should be able to handle fairly easily, freeing up time for executives to do more mentoring,” said Furlonger, a Gartner analyst whose survey recently included questions about AI. “CEOs are suggesting in the data that we don't need as many as we used to.”

The value of inexperience

Gil Luria, an equity analyst who covers Microsoft for investment bank DA Davidson, said one reason for the layoffs was that companies like Microsoft and Google were cutting costs to shore up their profit margins while investing billions in chips and data centers to develop AI. But another reason is that software engineers are susceptible to being replaced by AI at all skill levels, including experienced engineers who earn large salaries but are reluctant to adopt the technology.

Microsoft “can quickly do the math: see who’s adding value, who’s overpaid, who’s not overpaid, who’s adapting well,” Luria said. “There are veterans who have figured out how to take advantage of AI, and veterans who insist AI can’t write code.”

Harper Reed, chief executive of 2389 Research, which is building autonomous AI agents to help companies perform various tasks, said the combination of higher wages and a reluctance to adopt AI would likely jeopardize the jobs of experienced programmers.

“The way to cut costs isn't by firing the cheapest employees you have,” Reed said. “You take the cheapest employee and make them worth what an expensive employee is worth.”

YOU MAY BE INTERESTED IN: How to organize your digital library

Several studies have suggested this is possible. In a recent study conducted by researchers at Microsoft and three universities, an AI coding assistant appeared to increase the productivity of novice developers substantially more than it did their more experienced colleagues.

Reed said that, from a purely financial perspective, it would increasingly make sense for companies to hire entry-level AI employees to do what was previously mid-level work, with a handful of experienced employees to oversee them and almost no mid-level employees. That, he said, is basically how his company is structured. c. 2025 The New York Times Company