AI could disrupt work in developing countries before the productivity gains arrive

The latest predictions about artificial intelligence and jobs contain what looks like good news for developing countries. The World Bank’s 2026 World Development Report estimates that only 4.5% of jobs in developing economies are at risk of automation from generative AI, compared with 14.2% in high-income countries. At the same time, it estimates that AI could meaningfully improve productivity in 16.2% of jobs in developing economies.
So developing countries appear to face less automation while retaining much of the potential productivity gain.
But there is a problem with this picture.
The same conditions that make many jobs in poorer countries harder to automate can also make it harder for workers and businesses to benefit from AI. Weak electricity supply, poor connectivity, limited access to digital systems and the way work is organised can all slow the conversion of AI into higher productivity.
A joint study by the International Labour Organization and the World Bank describes this as the risk of “disruption without dividend”. Its analysis of 135 countries found that developing economies could experience changes to work before many of the benefits from AI become available to them.
That distinction matters.
Much of the debate about AI in developing countries still asks whether the technology will destroy jobs. That is too narrow. The more immediate problem may be that AI changes competition faster than it changes productivity.
A graphic designer in Nairobi, a software developer in Lagos or an accountant in Dhaka does not compete only with people in the same city. Digital platforms and remote work have already created international labour markets for many services.
Now imagine that workers and firms in richer countries begin using AI to complete some of those tasks faster and at lower cost.
The worker in Lagos may have access to the same ChatGPT or other AI tools. But access to the tool does not mean access to the same productive environment.
The AI is available, but the productive system around it is not. In my research with digital workers and small businesses in Nigeria, I have repeatedly seen the difference between having access to technology and being able to use it productively.
A business may own smartphones, use cloud services and receive digital payments, yet still organise parts of its work manually because electricity is unreliable or different systems do not connect properly. Workers find ways around these problems. They use backup power, several internet providers, multiple payment systems and combinations of digital and manual processes.
The business is digital. But using technology remains more costly and complicated than it should be.
AI enters this environment.
Take two small professional-services firms, one in Britain and one in a developing economy. Both can subscribe to the same generative AI service.
The British firm may already have reliable broadband, digitised records and established software for managing customers and projects. Its staff can use AI inside workflows that are already digital.
The other firm may have its information spread between WhatsApp messages, spreadsheets, paper documents and several applications that do not communicate with each other. Electricity and internet access may also be less reliable.
Giving both firms access to the same AI does not produce the same productivity gain.
The difference is not the intelligence of the software. It is everything around it.
This is why the World Bank’s finding that 16.2% of jobs in developing economies could benefit from AI should be read as a possibility, not an automatic outcome. The report itself stresses the importance of electricity, connectivity, skills and institutional capacity.
There is another problem.
Some of the workers most exposed to international competition are already working through digital markets.
A copywriter in Ghana may serve clients in Europe. A designer in Kenya may find customers through a global freelancing platform. A programmer in India may work remotely for companies in several countries.
AI does not need to eliminate these occupations to disrupt them.
A company that once hired five freelance writers may use AI and hire two people to edit and check the output. A small firm may decide that junior coding tasks can be completed by a more experienced developer using AI. Clients may expect the same freelancer to produce more work for the same fee because they assume AI has reduced the effort required.
The result can be pressure on earnings even when the occupation itself continues to exist.
There are already signs that AI’s effects may fall unevenly across workers. An ILO review published in June 2026 found that large-scale employment losses remain limited so far, but it pointed to concerns about weaker opportunities for younger workers in some exposed jobs and changes to worker autonomy and job quality.
For developing countries, this is important because digital work has been promoted as a route into global labour markets for young people.
If many entry-level tasks become cheaper to automate, the first rung of that ladder may become harder to reach.
Developing countries need a productivity strategy, not an AI adoption race. The wrong response would be to conclude that developing countries should slow down AI.
They cannot insulate themselves from a technology that businesses, clients and competitors elsewhere are already adopting. And there are real opportunities.
AI can help a small business prepare documents, analyse information or communicate with customers. It can give workers access to expertise that was previously expensive. The World Bank points to uses in areas such as healthcare, education and agriculture where relatively modest AI systems could extend scarce expertise to more people.
But governments should be careful about treating the number of people using AI as evidence of economic progress.
Downloading an AI application is adoption. It is not productivity.
The more useful question is whether AI allows workers and firms to produce more valuable goods and services, reach new customers, reduce costs or move into better-paid work.
That changes what AI policy should look like.
Skills matter, but teaching people how to prompt an AI system will achieve little if businesses cannot integrate the technology into how they actually work.
Connectivity matters too, but broadband by itself does not reorganise a firm.
Businesses need support to digitise records and improve basic management processes before sophisticated AI applications can do much with them. Workers need ways to demonstrate new skills and move into higher-value tasks as simpler work becomes automated. Small firms need access to finance so that productivity-enhancing technology is an investment they can actually afford.
Governments also need better information about which occupations are being affected. Counting AI users will not tell us whether earnings are rising or whether entry-level opportunities are disappearing.
This is where the debate about AI and development needs to change.
Developing countries may indeed face less direct automation than richer economies. But low exposure to automation should not be mistaken for protection.
If firms in richer countries become more productive while workers and businesses elsewhere struggle to make the same technology useful, the gap between them can widen without a dramatic wave of job losses.
The worker still has a job. The firm still exists. They are simply becoming less competitive.
That is the real risk behind disruption without dividend.
For developing countries, the question is no longer whether they can access AI. Many already can.
The harder question is whether their workers, firms and institutions can turn that access into higher productivity before the technology changes the markets in which they compete.