The central question is who captures the gain
An economy can become more productive while many households become less secure. Job counts alone miss wages, hours, bargaining power, ownership, prices, taxes, and access to essential services. AI’s distributional effects depend on institutions as much as technical capability.
Present evidence comes from generative-AI exposure, early adoption, and labor studies. Claims about a post-AGI economy are conditional forecasts. No dataset can reveal how income will be distributed after a technology that has not been established to exist.
Exposure is unequal
The ILO–NASK 2025 index estimated that one quarter of global employment had some generative-AI exposure and 3.3% fell in the highest category. Exposure was higher in high-income countries and in clerical work, and women’s employment was more exposed because occupations are gender-segregated (ILO–NASK, May 2025).
Exposure is not displacement. A highly exposed professional may receive a productivity-enhancing tool, while a less exposed contract worker may face tighter algorithmic monitoring. Workers with education and authority may be able to use AI to complement their judgment; workers in standardized roles may have tasks automated and discretion reduced.
The same technology can compress or widen wage differences. If it helps novices reach expert-like performance, employers may hire more entrants or pay less for scarce expertise. If it amplifies top performers with capital, data, and clients, returns may concentrate.
Labor income and capital income
Owners of chips, clouds, models, data, and distribution can receive returns even when employment remains stable. IMF staff modeling warns that stronger complementarity with high-income workers could increase labor-income inequality and that higher capital returns could increase wealth inequality. The authors state that these are staff views and scenario results, not the IMF Executive Board’s forecast (IMF Staff Discussion Note, 2024).
Ownership is therefore a policy variable. Pension funds and broad equity ownership distribute some capital income, but asset ownership is highly unequal and varies by country. Public investment, taxation, employee ownership, cooperatives, sovereign funds, and social dividends offer different ways to share returns. Each has governance and incentive tradeoffs.
Entry-level pathways may be the hidden fault line
Many occupations train experts through routine junior work: draft documents, basic analysis, customer questions, code maintenance, and supervised practice. Automating those tasks can reduce low-value labor while also removing the apprenticeship ladder.
Employers can redesign entry-level roles around review, client context, and tool supervision, but that requires deliberate investment. If every firm expects another institution to train experienced workers, the sector can face a collective skill shortage. Schools and licensing bodies should track whether graduates obtain supervised experience, not only whether senior employment persists.
Evidence of reduced hiring in some exposed young-worker categories is an early signal, not proof of universal causal displacement. Interest rates, industry cycles, and post-pandemic hiring corrections are competing explanations. Cohort hiring, wages, promotions, and task content should be tracked over time.
Prices and demand can spread gains
Productivity can benefit households through lower prices or better quality even when wages do not rise. Cheap translation, software, tutoring, or professional assistance may expand consumption and create complementary jobs. In other markets, dominant firms may retain savings as profit instead of reducing prices.
Demand elasticity matters. If a service becomes cheaper and people buy much more of it, employment may grow despite automation. If demand is fixed, fewer workers may be required. Regulation, reimbursement, procurement, and trust influence this response in health, law, education, and finance.
Quality-adjusted measures are essential. A free automated service is not a benefit if errors go uncorrected or human alternatives disappear.
Geography and the global divide
AI activity clusters around research institutions, capital, data centers, and reliable power. Regions hosting infrastructure may gain construction and tax revenue while bearing electricity, water, noise, and land costs. Regions dependent on outsourced clerical or digital work may face income loss without receiving model ownership.
The IMF estimates higher near-term exposure in advanced economies but warns that lower-income countries may be less ready to capture benefits. Lower exposure is not protection if it reflects poor connectivity, skills, language coverage, and public capacity.
Local-language systems, affordable compute, interoperable public infrastructure, and participation in standards can broaden gains. Data extraction without compensation and imported systems unsuited to local institutions can deepen dependency.
Work quality and algorithmic management
AI may remove repetitive tasks, assist disabled workers, and improve safety. It can also intensify monitoring, set opaque targets, schedule workers unpredictably, and automate discipline. Employment and output statistics do not capture autonomy or dignity.
The ILO’s 2025 cross-regional case studies found that worker representation and social dialogue can influence decisions about algorithmic management and employment (ILO Working Paper 144, July 2025). Case studies demonstrate possible arrangements, not average global outcomes.
Workers need notice about monitoring and automated decisions, access to the data used, a meaningful explanation, human appeal, and protection for collective bargaining. Consultation after procurement is less effective than participation in design.
Policy options are choices, not predictions
Education and transition support can fund portable training, apprenticeships, wage insurance, relocation assistance, and career services. Training is not sufficient when jobs are absent or when workers cannot afford time away from work.
Income floors—unemployment insurance, tax credits, child benefits, basic income, or guaranteed services—differ in targeting, administration, work incentives, and fiscal cost. A future AGI dividend is not currently available revenue.
Work sharing can convert productivity into shorter hours, but workers need bargaining power or law to retain pay. Competition policy can prevent rents and switching barriers. Tax policy can capture some gains, but poorly designed automation taxes may discourage beneficial tools or be easy to redefine.
Public options and procurement can make AI capacity serve public goals. They require technical competence, transparency, and safeguards against surveillance or political patronage.
What to measure
A distribution dashboard should report employment, vacancies, entry-level hiring, wages by percentile, hours, involuntary part-time work, labor share of income, capital ownership, prices, productivity, workplace surveillance, regional investment, and outcomes by sex, race, disability, age, and education where lawful and ethical.
Company announcements and employer surveys have conflicts of interest. Firms may promote productivity or justify layoffs; labor organizations may emphasize worker risk; consultancies sell transformation services. Administrative data, preregistered studies, and transparent methods provide stronger evidence.
AGI scenarios
If systems could substitute for much cognitive labor, bargaining power might shift rapidly toward owners of models and infrastructure. Alternatively, public ownership, abundant services, new demand, or reduced working time could raise living standards broadly. Physical bottlenecks and human preferences could preserve valuable work.
None is automatic. “Post-scarcity” is a political-economic claim as well as a technical one. Distribution rules, property rights, taxation, public services, and democratic control determine whether abundance becomes accessible.
The practical conclusion is to govern the distribution while capability is developing. Measure more than jobs, protect entry routes and worker voice, broaden ownership and access, and build fiscal systems that can share gains. Waiting until displacement is obvious may leave institutions weaker precisely when adjustment is hardest.