Rachel Lipson calls for policies to help early-career workers, even if the AI jobs shock is limited. Also, Anthropic analyzes research on whether worker retraining can meet the moment, how degree apprenticeships from EDvance College help early childhood educators earn more, and an essay on why big workforce problems require small solutions. (Subscribe here.)

Reforms Worth Making

Artificial intelligence’s impact on the job market remains unclear, even as hiring continues to tighten for early-career workers. Yet some experts and policymakers say the time is now to strengthen connections between education and work, whether or not an AI shock is coming.

In a new paper, Rachel Lipson makes the case for a dedicated early-career strategy for the AI transition. Lipson, a CHIPS senior policy advisor during the Biden administration who holds several research fellowships, says the stakes are high for young people, who tend to struggle with long-term economic penalties after entering a constrained hiring market.

“Making education more responsive to the pace of change can only help as the economy shifts,” she writes in the paper, which was published by the Center for the Governance of Change at Spain’s IE University. “Countries that use the AI moment to tackle long-standing weaknesses in their education-to-work infrastructure will not regret it.”

Lipson lays out the evidence behind three competing views on AI and the current entry-level slowdown. Her read is that the data is inconclusive, a take shared by most experts. The hiring outlook for recent college grads and other younger workers continues to worsen, however.

Employment among workers ages 22–25 in highly AI-exposed occupations is 19% below where it would be if it had kept pace with employment among similarly aged workers in less exposed occupations, according to updated research by Erik Brynjolfsson and his colleagues at Stanford University’s Digital Economy Lab. The relative decline in employment was 13% last summer.

The research, which was published last week, could not establish whether AI is driving these changes in the job market or if the documented patterns will accelerate, stabilize, or reverse.

The costs of a weakened entry-level job market are not felt equally. Lipson cites research finding that less advantaged college graduates suffer deeper long-term hits to their wages. And AI-driven early-career unemployment could hollow out talent pipelines for companies and deepen political disaffection among young people while draining tax revenue to pay for growing public assistance costs.

Eric Holcomb, the former Republican governor of Indiana, has said a potential AI jobs shock should be treated like preparing for a natural disaster. Lipson says she shares that urgency. (Holcomb is co-chairing RAISE US, a recently launched nonpartisan organization focused on the AI transition that’s raised more than $500M. Lipson is an advisor to the group.)

Policy Solutions: The paper draws on past examples of interventions during down economies. It explores solutions that are organized around potential market failures that could constrain entry-level hiring in an AI shock. Lipson says policymakers must ask whether we’re dealing with a skills problem, a matching problem, a true lack of demand with too few entry-level jobs, or a price problem where young workers cost too much relative to expected productivity. 

For example, under the assumption that young workers aren’t prepared for changing jobs—a skills problem—the paper describes opportunities for expanding apprenticeship, work-based learning, and employer training incentives. Lipson points to a reform push in France that more than doubled participation in apprenticeship, to 1M, over just five years. That expansion largely occurred within the higher education system, with a significant per-apprenticeship government subsidy and the creation of industry councils.

“Current experiments in states in nursing and teaching can model a path to broader scale” for apprenticeship in this country, she says. “But we probably need new federal legislation.”

The paper also looks at how education providers can tap employer partnerships to incorporate more real work experience. Community colleges should be a big part of solutions to an AI shock, Lipson argues. She points to the role of state incentives for employers to deepen partnerships with the two-year sector, citing research for her forthcoming book. 

“I like these policies because they acknowledge the reality that it is often difficult for companies to collaborate with colleges, and they may need incentives to overcome that friction,” she says. “This is worth public investment, in part because once a community college department starts working closely with a flagship employer, it improves the overall quality of their offerings.”

Lipson includes risks and caveats with the policy levers cited in her paper. For example, supply-side training on its own cannot compensate for an AI-linked drop in entry-level hiring. Even expanding apprenticeship—one of the paper’s safest bets—includes potential downfalls. Governments might pay for training that would’ve happened anyway, she writes, and job-specific skills can become obsolete.  

Yet we may deeply regret not acting while we wait to see what will happen with AI and jobs, Lipson concludes. Likewise, the education-to-work pipeline was broken long before AI entered the picture, Lisa Larson, a former community college president who leads the Education Design Lab, wrote in response to Lipson’s paper.

By reimagining education and training systems, Lipson says, policymakers can help young workers make better transitions into the workforce, regardless of how AI scenarios unfold.

The Kicker: “If we act early and the shock proves limited, we have spent modestly on reforms worth making anyway,” she writes.

— Sponsored By —

On September 15, Business Roundtable will host the 2026 CEO Workforce Forum, exploring The New World of Work. Join for conversations with CEOs, policymakers, and other leading voices on strategies to support upward mobility and opportunity in the age of AI. 

Meeting the Moment

If AI does lead to widespread job displacement, boosting government support for retraining workers is one of the most popular policy solutions cited by economists, AI experts, and the general public. Yet many experts argue that we can’t retrain our way out of AI’s economic disruption, often pointing to lackluster results from U.S. job-training programs.

Anthropic reviewed the evidence on worker retraining from 56 randomized studies. Claude Opus contributed to the meta-analysis by extracting traits and impacts from those studies. The resulting 122-page report, released last week, was co-authored by David Roodman, an independent researcher, and Maxim Massenkoff, an economist at Anthropic.

Stand-alone job-training programs were the focus of the analysis, which didn’t include high school vocational tracks, apprenticeships, or community colleges.

The training covered in the studies sought to prepare mostly low-income and young jobseekers for positions such as nursing aides, IT support technicians, and welders. Those programs took an average of six months to complete and cost roughly $13K per participant.

Here are top-line findings from the analysis:

  • Job-training programs have small impacts. They increased employment by an average of about 2 percentage points and annual earnings by $800 per year.
  • Large-scale training programs have also been lackluster. For example, an evaluation of the youth-targeted federal Job Corps found essentially no impact.
  • Some sectoral training programs boost earnings by 10 times as much. A few have lifted pay by $5K–$10K per year by intensively screening applicants, involving employers in deciding what to teach, and tracking local trends in the demand for skills.

“It seems doubtful that we currently have programs capable of meeting the moment,” the report says. “The impacts are small enough that they would not leave a dent in a persistently high unemployment rate.”

Anthropic also looked at research on the federal Trade Adjustment Assistance program, which has offered aid to workers who lose their jobs or wages due to foreign trade or offshoring. This approach includes training, income support, relocation allowances, and help with job search.

Evidence on the effectiveness of TAA is limited, the report says. But it’s worth considering for people who are demonstrably fired because of AI. That’s because the program gives people time to recover from job displacement, so that the loss of a job or even an occupation doesn’t cascade into a worse catastrophe.

The report also emphasized results for successful sectoral training providers such as Year Up, Project QUEST, and Per Scholas, which it said stood out in the research. These programs “can contribute to a crash response to AI disruption,” the report found.

However, attempts to replicate sectoral training have often failed. The report says it can take years to cultivate relationships with employers and develop competence in tracking regional demand for skills. Anthropic also cautions that these programs are not designed for skilled professionals, who may need years of retraining to recover their previous incomes.

Even so, the report calls for governments and other funders to invest now in demonstrating, evaluating, and growing the most promising training programs, especially sectoral training.

“Whether or not AI disrupts the labor market, we will not regret having learned how to best help workers adapt,” it concludes.

Apprenticeship

Opinion

Why Big Workforce Problems Require Small Solutions

When solutions to labor shortages and turnover focus solely on patterns across an entire workforce, they fail to reach the individual worker who is just trying to figure out if they can make it to their shift. 

Open Tabs

Data Center Jobs
OpenAI has signed a deal for an AI hub in Ohio that will be one of the world’s largest. With backing from Nvidia and SoftBank, the data center could cost up to $500B and produce eight gigawatts of computing capacity. (Northern Virginia’s total capacity hit four gigawatts last year.) OpenAI says the project will create 35K construction jobs and 2,500 long-term opening roles. The company is partnering with unions and investing in workforce training.

AI Infrastructure
Major manufacturers such as Caterpillar and Cummins are pivoting to meet new demand by equipping AI data centers, reports Bob Tita for The Wall Street Journal. That demand has helped to drive manufacturing last month to its highest level since 2022. For example, Ford Motor is looking to data centers to repurpose excess EV-batter production capacity. However, fears persist about the sustainability of the massive AI infrastructure build-out.

New Accreditors
The U.S. Department of Education released proposed rules to reform accreditation this week. The Trump administration’s goals for the regulation changes include simplifying the recognition of emerging and existing accrediting agencies. Nicholas Kent, the U.S. under secretary of education, has said that too few accreditors control access to billions in taxpayer funds and to professional licensure opportunities. The comment period for the rules will be open until Sept. 21.

Manufacturing Skills
Apple opened a new advanced manufacturing training center in Houston. The center is the company’s second learning site in the U.S. It offers free training and educational sessions to smaller businesses on how Apple products are made, including machine learning–driven quality control and advanced automation. The plan is for programming to expand at the center, which eventually will offer similar hands-on training to local college students.

Work-Based Learning
The State Higher Education Executive Officers Association launched a philanthropy-backed national coalition of 20 states and territories that are participating in a project to build sustainable, scalable approaches to work-based learning and quality education-to-career coaching. Eight of the states and territories will participate in a learning community and will receive structured assessment, technical assistance, and peer learning.

AI Applications
Reach Capital announced that it has closed on a new $265M fund. The San Francisco–based venture capital firm says the fund will back 50 early-stage companies. Its focus will be AI applications that can expand human potential across learning, health, and work, reports Dominic-Madori Davis for TechCrunch.

Job Moves
Eshauna Smith has been hired by StriveTogether as the first president of the nonprofit network of place-based partnerships. Smith has been director of community impact at the Ballmer Group since 2021 and was CEO of the Urban Alliance before that. 

Louis Soares has been hired by Axim Collaborative as the philanthropy’s VP of partnerships. Soares worked for 11 years at the American Council on Education and was the group’s chief learning and innovation officer.

Thanks for reading. If this newsletter was forwarded to you, subscribe here. —PF