Workforce Training and Reskilling Investment

CSET research emphasizes that workforce training works best when tied to sectoral training programs that bring together state agencies, local workforce boards, and employers within a single industry to develop needed talent. Empirical evidence shows these co-designed employer partnership models lead to better wage outcomes for workers compared to traditional classroom-only approaches, though they remain difficult to scale and often require participants to meet baseline skill requirements like high school diplomas that exclude the most vulnerable workers.

Revana Sharfuddin identifies six key restrictions in the U.S. Internal Revenue Code that effectively penalize human capital investment relative to automation. She proposes a suite of reforms to "restore neutrality," including eliminating the "new trade or business" bar (which prevents deducting training for new roles), abolishing the $5,250 annual cap on tax-free educational assistance, and removing the 5% owner limitation that blocks small business owners from accessing tax-advantaged training. 

Her central recommendation is to extend full and immediate expensing to all job-related training (mirroring the 100% bonus depreciation currently available for AI systems) thereby removing the fiscal incentive for firms to replace workers rather than retrain them.

Abigail Ball and Oren Cass propose "The Workforce Training Grant", which would redirect existing higher education subsidies to fund direct $10,000 annual grants to employers who create formal trainee positions. Unlike traditional retraining that funds the provider, this model funds the employer to create "paid apprenticeships," ensuring training is tied to actual labor demand. While they propose funding this by shifting higher education dollars, Anthropic’s policy team notes that taxes on AI consumption could also serve as a funding mechanism for these workforce development initiatives.

Gaspard proposes a "jobseekers allowance" that allows vulnerable workers to undertake training while still employed or immediately upon displacement without losing income. This shifts the model from reactive "dislocated worker" support to proactive adjustment, preventing the long-term unemployment spells that often render retraining ineffective. He also advises policymakers explore new approaches to retaining workers and leveraging “on-the-job training” (OJT) contracts under the workforce system to subsidize the wages of incumbent workers who need significant retraining.

The report calls for modernizing workforce training to meet the demands of the AI-driven economy by renewing the Workforce Innovation and Opportunity Act (WIOA) and extending Federal Pell Grants to cover high-quality short-term training programs. It also recommends expanding outcome-based "talent finance" models (such as income-share agreements and outcome-based loans) that allow workers to access quality programs without fully bearing upfront costs and risks.

The AI Workforce PREPARE Act mandates a study on designing more effective rapid retraining programs for workers displaced by AI. The legislation requires using AI-informed labor market projections to update states' "in-demand occupation lists," which determine how workforce training funding is allocated under existing programs. The bill also updates the Worker Adjustment and Retraining Notification (WARN) Act to require employers to provide basic information to employees if AI is a substantial factor in a layoff.

An initiative under the White House's America's AI Action Plan explicitly authorizing workforce agencies to use WIOA funding streams to build AI literacy pathways and training programs. The initiative emphasizes making AI training accessible through local career centers, developing stackable credentials and short-term certifications, and maintaining equity by building supportive services like childcare and transportation assistance into AI training programs to reduce barriers for underrepresented groups.

Writing for Brookings and AI Policy Perspectives, Jacobs argues that policymakers should be skeptical of retraining as the primary means of supporting labor adjustment to AI-driven automation, given decades of mixed-to-negative evidence from U.S. programs. He proposes four reforms: (1) developing better labor market projections by having AI labs work with policymakers to create granular forecasts drawing on capability evaluations and real-time job market data, (2) experimenting with AI-enabled hybrid and online training formats rather than the current 80% in-person model, (3) collating better evidence through RCT-style evaluations specifically targeting technology-displaced workers, and (4) reconsidering whether employment should remain the sole goal of training, instead exploring how programs might help people contribute to community life and social cohesion in scenarios where good jobs become scarce.