The reskilling versus upskilling debate is real, the distinction is useful, and most of the policy and corporate discussion treats both interventions as if they had similar success rates. They do not. Decades of randomized and quasi-experimental evaluations make clear that sectoral reskilling programs with employer co-design, named credentials, and direct job placement produce durable wage gains. Generic upskilling programs without those structural features produce essentially zero impact on earnings. The serious case in 2026 is not "we should do more of both." It is "let's stop calling things that do not work by the same name as things that do."
The argument here: the U.S. is spending roughly 0.1% of GDP on active labor market policies (compared to an OECD average of 0.5% and 1%+ in Denmark and Sweden), and within that small budget, much of the spend goes to interventions with weak evaluation evidence. Fixing the workforce-development infrastructure is less about scaling general "learning" and more about replicating the small number of program designs that produce measurable wage outcomes.
Defining the terms honestly
The standard definitions: reskilling teaches a worker the skills for a different job (a retail manager learning data analysis to move into supply chain). Upskilling teaches additional skills for the same job (a software engineer learning a new programming language). The distinction matters because the empirical evidence for each is quite different.
Reskilling, when it works, works because of structural features — not because of the curriculum
The randomized evaluation literature on U.S. workforce reskilling has converged on a clear picture. MDRC's long-running evaluation of Project QUEST in San Antonio, the gold-standard study in this space, found $5,000+ annual earnings gains that persisted for 11 years — among the largest documented impacts of any workforce program in the U.S. Per Sigelman and colleagues at the Burning Glass Institute have documented similar gains from Year Up, Per Scholas, and Project QUEST replications.
The mechanism is not the training content. It is the structural design: employer co-design of curriculum, paid training (or stipends covering opportunity cost), wraparound support (childcare, transportation, navigation services), named industry credentials, and direct job placement at the end. Generic online courses without those features — even when the curriculum content is comparable — produce significantly weaker earnings effects.
Upskilling has weaker but real evidence
Upskilling is harder to evaluate causally because most upskilling happens inside firms and its effects show up in productivity rather than wages. Erik Brynjolfsson, Danielle Li, and Lindsey Raymond's NBER paper "Generative AI at Work" (2023) — a quasi-randomized study of 5,179 customer-service agents — found 14% productivity gains overall and 34% for the lowest-experience-quintile workers using AI assistance. That is upskilling working: tools deployment plus training raises productivity at the bottom of the experience curve.
But the upskilling investment from U.S. employers has been falling for decades, not rising. Peter Cappelli at Wharton has documented this across his career; the OECD's Employment Outlook 2024 confirms it cross-nationally. The "skills revolution" rhetoric outruns the actual employer investment.
Where the U.S. workforce policy is failing
Public ALMP spending is one-fifth of OECD average
The most important number in this essay: the U.S. spends roughly 0.1% of GDP on active labor market policies (training, employment subsidies, job-search assistance), compared to 0.5% OECD average and 1%+ in Denmark, Sweden, and the Netherlands. That delta is the gap between countries where reskilling is routine and countries where it is heroic. The 2021 Bipartisan Infrastructure Law, the 2022 CHIPS and Science Act, and the 2022 Inflation Reduction Act have driven meaningful expansion of Registered Apprenticeships and sectoral training partnerships, but the U.S. starting point was low enough that even substantial growth leaves the system underfunded by international comparison.
Public WIOA evaluations show modest effects
RAND Corporation's evaluations of the Workforce Innovation and Opportunity Act and earlier Workforce Investment Act programs have found modest earnings impacts — typically 5–10% wage gains, sometimes fading after three years. The reason is well-documented: programs are too short, employer connection is weak, completion rates are mediocre, and most do not include the structural features (employer co-design, named credentials, direct placement) that distinguish high-impact programs like Project QUEST. Funding mass general training is less effective than concentrated investment in fewer programs with stronger design.
Employer training has declined for decades
Cappelli's research and BLS data both show that U.S. employer-funded training is materially lower as a share of payroll than in the 1970s. The WEF's Future of Jobs Report 2025 found that 86% of employers expect AI to transform their business by 2030, but only about half provide employer-sponsored AI training. The gap between expectation and investment is the empirical core of the upskilling failure.
What workers can do given the institutional shortfall
If your employer is not investing in training and the public workforce system is not adequately funded, the practical reskilling/upskilling strategy is structurally limited. Three moves with measurable evidence:
Pick credentials with employer recognition. Industry certifications (AWS, Cisco, CompTIA, Google Career Certificates, state professional licenses) produce 10–30% wage premiums over comparable workers without them. Generic MOOCs without named recognition produce near-zero wage premium. The signal is in employer acceptance, not curriculum quality.
Apply to Registered Apprenticeships. The U.S. Department of Labor's Registered Apprenticeship system has expanded substantially since 2021, and apprenticeship completers earn an average starting wage of roughly $77,000 — substantially above the median for the credential level. The system is now open to adult learners and mid-career switchers, particularly in construction, manufacturing, healthcare, and clean energy.
Pair domain experience with one current AI tool credential. The Brynjolfsson-Li-Raymond research and subsequent BCG and Stanford studies are unambiguous: AI tools deliver the largest productivity premium to workers who pair them with existing domain expertise. Microsoft Copilot, Salesforce Einstein, AWS, or Google AI certifications are inexpensive, fast to earn, and the most directly relevant skill upgrade in the 2026 labor market.
What policy should actually do
Three reforms with measurable evaluation evidence:
Scale sectoral training to match Switzerland-level per-capita apprenticeship coverage. The U.S. is roughly an order of magnitude below the Swiss, German, and Austrian per-capita rates. Robert Lerman at Brookings and Annelies Goger at the same institution have published detailed proposals for how to scale.
Co-fund employer training as a refundable tax credit, tied to evaluation outcomes. Existing federal tax incentives for employer training are weak and unconditional. Conditioning the credit on documented completion and wage-gain outcomes would shift incentives toward programs that actually work.
Expand portable training accounts for workers in industries undergoing transition. The EU's Platform Work Directive (2024) attached portable training credits to gig workers — a policy mechanism the U.S. has yet to seriously consider.
For the broader argument on what reskilling looks like in 2026 — and which program designs actually deliver durable wage gains — see NWLB's Reskilling for Real →.
The reskilling-versus-upskilling debate hides the more important distinction: between programs that work and programs that do not. The U.S. has data on which is which, and chooses to fund both at scale anyway.
"Re-skilling vs. up-skilling" gets framed as a strategic taxonomy question. The more important question is which programs in either category actually produce measurable wage gains, and the answer is unambiguous in the evaluation literature. Workers who navigate the current institutional shortfall successfully are mostly the ones who treat reskilling investment like a portfolio: a small number of high-recognition credentials, an apprenticeship pathway where available, and a deliberate pairing of domain experience with current AI tools. The institutions should be funding more of those at scale. They are not, yet.
Updated May 21, 2026. This piece was substantively rewritten as part of NWLB's 2026 editorial refresh.
