Africa
The risk is exclusion from the gains, not displacement.
26%of jobs exposed to AI in low-income countries (IMF)
Africa playbook →Regions · Last updated November 2026
Direct answer
AI can create new jobs in every developing region, but the starting point differs. In Africa the risk is exclusion from the gains. In South Asia it is the outsourcing jobs that AI replaces first. In Southeast Asia it is reaching the informal majority, in Latin America the digital gap, and in the Middle East the distance between Gulf investment and a young workforce short of work. Applied Frontier publishes a playbook for each region and country reports that are updated every year.
Five regions, one structure: how exposed the region is, why the next three years matter, which job families to build first, what a ministry should do this year, and where multilateral agencies fit.
The risk is exclusion from the gains, not displacement.
26%of jobs exposed to AI in low-income countries (IMF)
Africa playbook →The outsourcing and IT jobs that built the export sector are the ones AI replaces first.
15%of jobs strongly complementary to AI (World Bank)
South Asia playbook →Most exposure is in work AI can assist, but nearly 80% of workers are informal.
22.9%of employment in occupations exposed to generative AI (ILO)
Southeast Asia playbook →The gains are real, but about half are blocked by the digital gap.
17Mjobs that could gain from AI but lack digital access (ILO and World Bank)
Latin America playbook →Gulf investment on one side, a young workforce short of work on the other.
26.2%youth unemployment in the Arab States in 2025 (ILO)
Middle East playbook →Country reports · 2026 edition
Kenya, India, Bangladesh, Indonesia and Brazil. In each, training for an AI-economy role costs roughly one-eighth to one-sixteenth of a four-year degree, on Applied Frontier’s benchmark. The benchmark covers training only; targeted pilots measure the full cost per placed worker.
Africa · 2026 edition
800k–1Mlabor-market entrants a year
Benchmark training cost of an AI-economy role: $760–$1,510
Kenya country report →South Asia · 2026 edition
7.85Mnon-farm jobs needed a year
Benchmark training cost of an AI-economy role: $240–$480
India country report →South Asia · 2026 edition
2Mlabor-market entrants a year
Benchmark training cost of an AI-economy role: $340–$680
Bangladesh country report →Southeast Asia · 2026 edition
1.86Madded to the labor force in a year
Benchmark training cost of an AI-economy role: $460–$930
Indonesia country report →Latin America · 2026 edition
8.1Myoung people outside work and study
Benchmark training cost of an AI-economy role: $2,440–$4,870
Brazil country report →| Country | Jobs challenge | Benchmark training cost, AI-economy role | Cost of a degree |
|---|---|---|---|
| Kenya | 800k–1M labor-market entrants a year | $760–$1,510 | $12,105 Four-year computing degree |
| India | 7.85M non-farm jobs needed a year | $240–$480 | $3,805 Four-year engineering degree, minimum fee |
| Bangladesh | 2M labor-market entrants a year | $340–$680 | $5,426 Four-year public-university degree |
| Indonesia | 1.86M added to the labor force in a year | $460–$930 | $7,400 Four-year computer science degree |
| Brazil | 8.1M young people outside work and study | $2,440–$4,870 | $38,986 Four-year federal-university degree |
Benchmark: public teaching cost per student applied to a program of three to six months. Sources and method are in each report. Figures for Kenya, India, Indonesia and Brazil are converted at 2024 average exchange rates and Bangladesh at the September 2026 rate.
Eight families that sit between a global model and a local outcome. None requires a four-year degree. All can be trained in weeks to months, and most cannot be offshored. Each regional playbook names the four to build first.
3–6 months to train
Installs, configures, secures and maintains AI systems inside businesses, clinics, farms and government offices. Local, hands-on, in chronic shortage everywhere.
2–4 months on top of domain experience
Checks AI output in health, agriculture, finance, law and education before it is acted on. The model is global; the verification is local.
2–3 months to train
Collects, labels, governs and licenses data in languages and domains the models lack. The asset is the data, not cheap annotation.
2–3 months to train
Assembles AI agents into working processes for small firms: invoicing, inventory, customer service. Millions of SMEs with no software.
2–6 weeks added to an existing role
Community health worker, extension agent, teacher or clerk operating with an AI copilot. The largest employment pool in any developing economy.
3–6 months to train
Sensors, drones, solar and storage, logistics, with AI in the loop. Energy and agriculture are among the largest sectors in most developing economies.
3–6 months to train
Data-center operations, power, cooling, network. Follows the compute investment arriving in 2025–27.
6–12 months with management experience
The accountable human for an AI-run process in a bank, ministry or utility. Liability and trust remain human.
Not on the list as a strategy: data annotation, entry-level software development, “prompt engineer” roles and call-center work. Annotation is a task we will place in the best market for it; it is not a national strategy, because it is priced by the buyer and being automated.
In three levels. Each region has a playbook. Each country report sits under its region, and each report is updated every year with the same indicators, so that countries can be compared with each other and with their own earlier results.
Five regions
One page per region: exposure, the three-year window, priority job families, actions for ministries and the role of multilateral agencies.
Five countries
One permanent page per country, under its region. It always shows the latest edition.
2026 edition
Country data is updated every year with the same indicators, so progress can be compared year on year.