The Invisible Export

Executive Summary AI will reshape job descriptions sector by sector but is unlikely to produce net, economy-wide job losses

GenAI’s labour-market effects will be uneven, with women more exposed to task displacement

The technology is advancing exponentially , not linearly, and a ‘Holy Cow’ moment may arrive sooner than anyone realises

Services is displacing manufacturing as the main engine of global trade

India’s services advantage is real, but its English-proficiency moat is weakening

AI-enabled services offshoring could narrow global income gaps as EM workers become more productive and globally competitive

. Since 1990, the global services trade has grown at roughly twice the pace of the goods trade. This gap will only widen as generative AI erodes barriers that have long protected high-wage service jobs in advanced economies: the cost of transferring skills, and the requirement of knowing a common language. For India, where business and professional services are already among its fastest growing exports, this shift might cut both ways. It could define the next phase of export-led growth, but it may also reshape its competitive advantages.

Richard Baldwin examined how AI is rewriting the rules of competition and what this shift means for jobs, trade strategy and the next phase of globalisation. He framed the discussion as a set of provocations rather than predictions, using these to test common assumptions about AI, work and the next phase of globalisation.

Provocation 1: The Future Is Not Fewer Jobs Professor Baldwin’s first provocation challenged the most common framing of AI disruption: that AI will destroy jobs. While he conceded that some jobs may be displaced at the sector or occupation level, at the same time, other sorts of jobs – even within the same sector/occupation – may get created or scaled up. However, at the economy level, AI’s likeliest outcome is not permanent job losses, but large-scale labour-market restructuring. Individual sectors may experience severe disruption, but aggregate employment will adjust, as it always has, as workers move across sectors and forms of work.

The broader question is not whether AI will destroy jobs, but where it will tip the balance between productivity gains and demand expansion. Technological progress reduces the number of workers needed per unit of output, but it can also lower prices, improve quality and thus lift demand by enough to offset the impact of having fewer workers per unit. In the 19 th century, mechanisation pulled workers into manufacturing and made goods cheaper; on net, the demand-push effect outweighed the effect of productivity gains. From the late 20 th century on, information technology pushed workers out of manufacturing and into services, but aggregate employment did not fall. The same lens applies to AI. The disruption it will create will be uneven and often painful, but jobs will not disappear. Rather, tasks, workflows and skill requirements will change. In this sense, AI will not primarily destroy jobs; it will destroy job descriptions .

Provocation 2: Women Will be Hit Harder by AI Job Displacement A 2026 International Labour Organisation study argues that the tasks Gen AI performs best, such as copy-editing, translation and routine formatting, are highly concentrated in occupations where women are overrepresented. The finding rests on a straightforward methodology: assessing which tasks which are most readily automated and the occupations in which those tasks cluster. The results suggest that jobs that are more typically ‘female dominated’ tend to have a higher share of ‘automatable content’ than those that are male dominated. The issue is not differences in gender capability , but the current distribution of jobs across genders at an economy level. The study implies that AI’s effects on the labour market will be unevenly distributed by gender, whatever its broader, economy-wide effects may be.

Provocation 3: Beware of the ‘Holy Cow Moment’ The third provocation rests on what is sometimes called Amara’s Law: the tendency to overestimate a technology’s short-run effects and underestimate its long-run ones. Each new wave of technology creates excitement, anxiety, or both, but when the expected disruptions do not immediately arrive, people get reassured, even complacent. This shift from alarm to reassurance, though, is often quickly followed by a ‘Holy Cow’ moment, when expectations flip and businesses realise the technology has moved further even than their operating models assumed.

Gen AI’s ability for recursive self-improvement strengthens this risk. As AI systems begin improving agents and workflows through feedback loops, capability growth may no longer follow the straight-line assumptions on which businesses typically forecast change. The risk lies in mistaking a quiet early phase for evidence that disruption is still some distance away.

Provocation 4: The Future of World Trade Is Services Globalisation has seen three successive ‘unbundlings’, each made possible by falling transportation or communication costs. The first unbundling occurred around 1820, when the steam engine enabled goods to be transported more easily (and cheaply) cross borders, thus firmly separating production from consumption. The second came in the 1990s, when factories crossed borders, allowing a separation of stages of manufacturing, and combining advanced-economy technology with emerging-market wages. The third unbundling was accelerated by the pandemic, which separated offices from each other and allowed the same high-tech, low-wage combination to operate in services rather than goods.

Three factors suggest that the future of world trade will be built around services rather than goods:

Global goods exports have plateaued while services exports have kept growing . Since the Global Financial Crisis, the goods trade, as a share of GDP, has slowed, pulled down by rising geopolitical frictions, tariffs barriers and supply-chain localisation. Meanwhile, digitally-enabled B2B and B2C services exports have steadily grown.

EM are growing their share of the services trade faster than developed economies . A growing global services market favours countries with deep pools of educated labour and established delivery capability. This is a significant advantage for emerging markets like India. However, as capabilities get built across the globe, India will not remain the only credible low-cost services platform for long.

Manufacturing-export-led growth has become harder, partly because of China’s scale and competitiveness. Automation is reducing manufacturing jobs even where output is rising. Services, by contrast, may offer a longer runway for job creation, productivity gains and export growth. This does not make manufacturing irrelevant, but it does weaken the argument that it is the only viable development path. (This pattern is documented well in the World Bank’s report ‘At Your Service’.)

Provocation 5: Offshoring Maps Will be Redrawn Two sorts of technology are dissolving the remaining barriers to services offshoring: large language models (LLMs) and speech translation services. Together, they will lower the cost of diffusing skills, reduce language barriers and make remote service workers less remote.

LLMs as instant skill diffusion

LLMs radically lower the cost of transferring skills and experience to foreign remote workers. These models are trained on work done by the highest-skilled and most highly-paid professionals in advanced economies, then made available to anyone and anywhere, for the cost of a modest monthly subscription. This changes the economics of offshoring. If a human still needs to remain in the loop, it may become more viable to place that human in a lower-wage country, provided quality and governance standards are met. The value of the worker then lies in the ability to use AI tools well and exercising judgement. In this model, AI does not remove the offshore worker, but raises the range of tasks that can be offshored.

Professor Baldwin’s own research with online freelancers points in this direction. Workers from an advanced economy and an emerging economy were asked to complete the same set of tasks, with and without LLM support. Without ChatGPT, the differences in quality and productivity were apparent. When both sets of workers used the tool, however, quality gaps narrowed considerably and evaluators found it harder to distinguish between output from the two geographies. LLMs therefore not only diffuse skills, but also narrow differences in business style, written presentation and cultural fluency, that would otherwise shape perceptions of work quality.

Falling language barriers

Language has long acted as a barrier to trade, including the services trade. For India and the Philippines, English proficiency has been a major advantage in business process outsourcing. This advantage is now becoming less defensible as real-time speech translation improves and becomes more widely embedded in consumer and enterprise tools. Countries with large pools of skilled workers but weaker English-language capability, including parts of Latin America, Southeast Asia, Africa and China, could become more viable services exporters with the aid of speech translation technologies.

Provocation 6: GenAI Will Reduce Global Inequality As LLMs narrow the service-sector skill gaps between emerging and advanced-economy workers, productivity gaps will also narrow. Wages tend to adjust more slowly than productivity does, but these gaps are what makes EM service offshoring increasingly attractive to companies in advanced economies. The resulting boom in service-sector offshoring will narrow income gaps between nations, forming the sort of services-led middle-class that Professor Baldwin terms ‘Bangalore’s Galore’. He expects this dynamic to extend beyond India and China, to cities as widespread as Nairobi, Bogotá, Quito and Cape Town, where sizeable pools of skilled labour could become globally competitive as GenAI narrows capability gaps while wage differentials slowly adjust.