decade, and well below the figure for Americans overall, at more than a quarter. A slump in on-campus recruiting is not helping. Job postings on Handshake, a search platform for university students, are 50% below their peak of 2022. Many suspect AI is to blame. More than half of employers say they have considered replacing entry-level workers with the technology. A recent poll by the Institute of Politics at Harvard University’s Kennedy School found that, similarly, more than half of young Americans see AI as a threat to their job prospects. Economists are more divided. A paper by Erik Brynjolfsson of Stanford University and colleagues, published in 2025, examined employment among young workers in AI-exposed jobs, such as software development. The authors found it had fallen by 16% relative to less-exposed fields. But a paper published this year by Zanna Iscenko and Fabien Curto Millet, two economists at Google, casts doubt on the idea that young workers in particular are being displaced by AI. They found that job postings in AI- exposed occupations have declined just as sharply for senior workers as for junior ones, and that this trend predated the launch of ChatGPT in late 2022. Another study, by Morgan Frank of the University of Pittsburgh and colleagues, showed that labour-market outcomes have deteriorated for AI- exposed employees, but that this trend also began before ChatGPT was released.
The Economist has conducted its own analysis, using a largely overlooked source of data: ten years’ worth of surveys of recent college graduates from the National Association of Colleges and Employers. Each year American universities ask new alumni whether they are working, unemployed or in graduate school. Using their responses, we compared labour-market outcomes in fields with differing levels of exposure to AI before and after the arrival of large language models. We found that graduates in fields more exposed to AI have suffered markedly worse outcomes. Between 2022 and 2024 graduates in the least- exposed quintile—studying subjects such as education, philosophy and civil engineering—saw their average full-time employment rate fall by just 1.5 percentage points. Those in the most exposed quintile—including computer science, computer engineering and information science—suffered a 6.6 percentage-point drop (see chart 1). We updated these figures for the most exposed fields, using data from 13 universities, and found that the trend continued for the class of 2025 (see chart 2). The rate of full-time employment fell from nearly 70% to 55% in three years—notably, the three years following ChatGPT’s release in 2022. Prior to that, it had been stable.
Students are already changing course. Data from the National Student Clearinghouse, a research group, show that undergraduate enrolment in computer science fell by 11% in 2025. Enrolment in computer programming, which focuses on coding skills rather than theory, dropped by 26%. The work done by computer-science graduates is changing, too. Less time is spent writing code; more is devoted to designing and organising software systems at a higher level. Lana Yarosh, director of undergraduate studies in computer science at the University of Minnesota, says she understands students’ anxieties. “It’s always hard when things change. But computer science is a field where everything changes all the time.”■ For more expert analysis of the biggest stories in economics, finance and markets, sign up to Money Talks, our weekly subscriber-only newsletter. This article was downloaded by zlibrary from https://www.economist.com//finance-and-economics/2026/05/13/is-ai-putting- graduates-out-of-work-already
Finance & economics | (Robots) serve the people China wants more robots but not fewer workers A human-first approach to automation May 14th 2026 A year ago the city of Qingdao had just a handful of autonomous vehicles. Now it has more than almost anywhere else on Earth. One firm, Neolix, has put around 1,200 unmanned delivery vans on local roads; it hopes to have 4,000 by the end of the year. With several other autonomous taxi and food- delivery projects under way, Qingdao exemplifies how rapidly artificial intelligence is transforming China. It is also the front line of the clash between unmanned vehicles and drivers. Autonomous cars and drones are being deployed in China at a dizzying pace. About 33,000 short-range delivery vehicles, including the ones in Qingdao, were on Chinese roads at the end of 2025. The number of unmanned cabs is expected to hit 14,000 by the end of 2026. Goldman
Sachs, a bank, reckons that more than 700,000 robotaxis (meaning 12% of all ride-hailing vehicles) will roam Chinese cities within five years. Meituan, a delivery super-app, believes it could use drones for 10% of the country’s instant food deliveries, of which 60bn were made last year. Though each such delivery is a technological miracle, in the short run it may deprive a human driver of a fare. This puts Chinese leaders in a bind: they want to lead the world in AI and automation but not destroy jobs. An economic plan for the next five years says the country must “prevent and resolve large-scale unemployment risks”. In April a cyber-security watchdog told developers in a draft document that they should “not apply AI with the goal of replacing human employment”. The first question is whether technology will be able to replace millions of drivers in short order. Projects have been launched in dozens of Chinese cities, but growth has slowed owing to congestion and technical glitches. No company has deployed more than 1,200 vehicles in any one city. The size of Neolix’s fleet in Qingdao has fluctuated in response to traffic jams caused by its vans. Although these are permitted to drive at any time, they are only allowed to make deliveries only in off-peak daytime hours. Even so, on an April morning near a large wholesale market, bands of Neolix vehicles could be seen clogging up the road to honks and jeers. In the city of Wuhan, home to one of the world’s biggest robotaxi projects, autonomous vehicles have also jammed traffic. Baidu, a tech giant and Wuhan’s main operator of driverless taxis, has a fleet of around 1,000 for well over a year. It may not grow for a while. In March dozens of Baidu’s cabs suddenly froze, snarling up traffic and prompting a rescue effort for stranded passengers. Since then the central government has suspended the issuance of new licences for robotaxis. A second question is which jobs are threatened and which are safe in the longer run. Authorities in Qingdao are not worried about unemployment caused by Neolix, says Wang Honglei, a company executive. In fact, senior officials in the wider Shandong province want as many as 15,000 driverless short-delivery vehicles on roads by the end of 2027. One reason for the insouciance is the sort of human drivers these might displace. Neolix runs only business-to-business services, such as delivering meat from markets to
restaurants. Many people who do this are in their 60s and drive small, three- wheeled vehicles that tip over in traffic. Few young people seek to replace them when they retire because this perilous work pays poorly and entails a lot of heavy lifting. This makes machines the obvious choice for the job. Drivers who ferry people to their destinations and packages to consumers are another matter. Tech platforms employ around 22m such workers; many more drive city cabs. The platform workers are generally young, rural migrants in cities and people who have lost other work. Youth unemployment is already high and officials do not want to make it worse. Another crucial difference is that taxi and delivery drivers have been better at organising strikes and protests. They did just that in Wuhan in 2024, as Baidu’s project was gaining momentum. In response city officials told Baidu to stop publicising its robotaxi figures. Authorities, though supportive of the technology, fear social disruption far more than tech glitches. To allay those fears, the biggest automation companies are providing help to the disaffected. Meituan has started training delivery drivers to help operate drone deliveries in Shanghai. Jobs range from loading food onto drones to monitoring flights from a command centre. So far only 200 people are on this team, compared with millions of drivers. But this will expand, says Mao Yinian of Meituan. Whereas the company previously only trained its own employees for such work, he says, it now trains other workers, too. These include staff at hospitals, where drones already deliver some test samples. The point of automation is ultimately to replace workers, who need to be paid a regular wage, with robots, which do not. This will eventually end up being true in China, too. In the meantime, a world anxious about an AI jobs apocalypse will be watching the Chinese experiment in human-first automation closely.■ For more expert analysis of the biggest stories in economics, finance and markets, sign up to Money Talks, our weekly subscriber-only newsletter. This article was downloaded by zlibrary from https://www.economist.com//finance-and-economics/2026/05/11/china-wants-more- robots-but-not-fewer-workers
Finance & economics | The Lazarus effect America is experiencing a productivity miracle AI hasn’t—yet—got much to do with it May 14th 2026 As with many a miracle, onlookers disbelieved their eyes at first. For a decade after the global financial crisis of 2007-09 rich-world productivity growth was, by historical standards, deadish. Since prosperity depends on the ability to produce more with the same labour, this consigned even America to eternal stagnation (and don’t ask about Europe). The Congressional Budget Office, a fiscal watchdog which consistently overestimated productivity growth in the 2010s, has been consistently glum this decade (see charts 1 and 2). Partial data hinting otherwise were dismissed as false prophets.
But those data kept coming. And now they are indisputable: over the past five years or so American productivity has been growing at the fastest rate in around two decades. Whether you look at output per worker or per hour, it has risen by a lively 2% a year, from a moribund 1% for most of the 2010s (see chart 3). This has led the Federal Reserve to raise its median forecast for America’s long-run GDP growth from 1.8% to 2%. Jerome Powell, the outgoing chair, bore witness at a recent press conference. “I never thought I’d see this many years of really high productivity,” he marvelled in response to a question from The Economist.