The phrase "digital divide" has done a lot of work over the past twenty years and is now actively misleading. The 2024 version of the divide is not about broadband access (97% of U.S. households have it, per Pew Research's annual internet use surveys); it is not about device ownership (more than 90% of working-age Americans own a smartphone). It is about something narrower and more consequential: which workers know how to use software stacks that produce a six-figure professional income, and which workers' jobs are mediated by software stacks that surveil and control them. The post-pandemic divide is not access. It is fluency, and it is on which side of the algorithm a worker sits.
Saying it that way changes the policy conversation. The investment that closes the digital divide in 2026 is not laying more fiber. It is teaching the specific applied software skills that produce wage premia, and protecting workers from the algorithmic management practices that increasingly run logistics, hospitality, retail, and gig work. The two halves of that agenda are usually treated as unrelated. They are the same agenda.
The fluency gap, in numbers
The OECD's Skills Outlook and the underlying PIAAC (Programme for the International Assessment of Adult Competencies) data are the cleanest international read on this. PIAAC scores adult problem-solving in technology-rich environments — essentially, can you accomplish a multi-step task using software you have not seen before. The 2023 cycle found that roughly one in three U.S. adults scored at or below Level 1, meaning they can perform only the simplest, single-step digital tasks. That share has barely moved since the 2012 baseline. Among older workers and workers without a four-year degree, the share scoring at or below Level 1 exceeds 50%.
Translate that to the labor market. The Brookings Institution's Digitalization and the American Workforce series, led by Mark Muro, has scored U.S. occupations on a digital-skill intensity scale and tracked how that intensity has changed over time. Their finding is that the share of jobs in the "high digital" category roughly doubled between 2002 and 2020, and the wage premium for high-digital occupations is now roughly 50–80% above low-digital occupations after controlling for education. The gap is not closing. It is widening, faster in some metros than others.
The algorithmic-management side of the divide
The newer fact, less visible in older "digital divide" discourse, is that low-wage work is increasingly digitally mediated in the other direction. Warehouse workers, ride-hail drivers, food-delivery couriers, retail associates with electronically scheduled shifts, and contact-center workers with productivity-monitoring software are all subject to what researchers call algorithmic management. Veena Dubal at UC Hastings has documented this in her work on "algorithmic wage discrimination"; Alex Rosenblat's Uberland (University of California Press, 2018) was the first widely-read account; the EU Platform Work Directive of 2024 is the first major regulatory response.
The empirical pattern is consistent: workers in algorithmically managed roles report lower schedule predictability, higher work-related stress, and lower wage growth than peers in non-algorithmically managed equivalents. The pandemic accelerated the rollout of these tools by lowering the social resistance to surveillance and by giving employers a "safety" framing for what was actually a productivity intervention. They have not been rolled back.
What actually closes the gap
The most rigorously evaluated interventions in this space have been sectoral training programs that combine intensive applied digital skills with explicit employer commitments to hire and promote. MDRC's evaluations of Per Scholas (a 13-week IT-skills program serving primarily Black and Latino adults without four-year degrees) found earnings gains of roughly $7,500 per year sustained at two and three years post-program, with effect sizes that put it among the most cost-effective workforce interventions ever evaluated rigorously. Year Up's randomized control trial, published in the American Economic Journal, found similar magnitudes. These programs work because they are sectoral, applied, and tied to specific job pathways — not because they teach abstract digital literacy.
The federal infrastructure bills passed since 2021 — the Bipartisan Infrastructure Law, the Inflation Reduction Act, and the CHIPS and Science Act — have made roughly $250 billion available for broadband, workforce development, and advanced manufacturing training between them. The early data on uptake is uneven; the GAO's 2024 reviews have flagged that workforce funding is reaching incumbent training providers more than new sector-specific programs. The next administration's implementation choices will determine whether these dollars produce another tranche of generic certificate programs or fund the sectoral models that have actually worked.
The AI overlay that changes the calculation
Generative AI raises the stakes on both sides of the divide. For high-skill knowledge workers, AI tools amplify productivity — the empirical literature here, including the GitHub Copilot productivity studies and the Brynjolfsson/Li/Raymond Quarterly Journal of Economics 2025 paper on customer-service AI, finds productivity gains concentrated among lower-tenure and lower-skill workers in those job categories. AI compresses the within-occupation skill gap. For workers in algorithmically managed jobs, AI is making the management denser and the surveillance more granular. The two-tier system is not slowing; it is accelerating along both axes.
This connects directly to the Who Gets Augmented, Who Gets Replaced → pillar, which traces this divergence in detail. The honest reading is that "digital divide" is now too coarse a phrase. The relevant divides are: who controls the algorithm versus who is controlled by it, who is augmented by AI versus replaced by it, and whose digital skills carry a wage premium versus whose digital exposure is a wage ceiling.
The reform agenda, briefly
Three things matter. First, fund the sectoral training models with evidence behind them — Per Scholas, Year Up, NPower, Apprenti — rather than spreading workforce dollars across generic certificate programs. Second, regulate algorithmic management directly: the EU Platform Work Directive's transparency and human-oversight requirements are the template U.S. policy should learn from. Third, treat digital-skill investment as continuing education rather than youth education: the BLS data on retraining-induced wage gains is consistently strongest for workers who retrain in their 30s and 40s, not their 20s. The age distribution of training spend should follow that finding.
The digital divide in 2026 is no longer about access. It is about which workers run the software and which workers are run by it — and the next decade of workforce policy will be judged on whether it narrowed that gap or built it in.
Updated May 21, 2026. This piece was substantively rewritten as part of NWLB's 2026 editorial refresh.
