Originally published here.
Two big forces are pulling at opposite ends of the same struggle over how to sustain productivity: Birth rates and AI automation. In the West, and increasingly elsewhere, birth rates are falling, which means the supply of new workers will shrink. At the same time, AI-driven automation is rising, which means the demand for human labour may also shrink or at least change shape. The central question is whether these curves intersect in a way that stabilises living standards and social contracts, or whether one force overwhelms the other, tipping society into imbalance and disruption.
Fertility keeps decreasing in rich economies. In the EU, total fertility fell to 1.38 births per woman in 2023. The replacement rate at which a population size is stable is 2.1. Driven largely by international migration, the EU population is expected to peak around 2026 at ca. 453m and then decline. By 2100, we will have “lost” ca. 30m people from the peak we once had in 2026 and only count ca. 420m people.
The labour force reacts with a lag. Births today become entrants to the workforce in roughly 20 years. That means the birth slump of the 2000s shows up as fewer new workers from the late 2010s through the 2020s and 2030s. The working-age populations will contract in 22 of 27 EU countries by 2050. The old-age dependency ratio is projected to reach roughly 57 percent by 2050, which is fewer than two workers per person aged 65+. Thus, the total amount of hours worked are projected to fall even if employment rates rise.
On the other hand, AI-driven automation is real but unevenly distributed. While AI already has significant impact on the creative industries and in the form of various chatbots, AI agent adoption varies by industry, company and use case. Some companies are successfully replacing legacy software through in-house agentic solutions, while for many real-world use cases, usability and thus adoption is still lagging. For example, despite several agentic browsers having launched, we cannot yet outsource our entire shopping to AI agents due to a lack of reliability as well as specific context that is required - and this data often isn’t readily available. Many agentic capabilities will take time to mature before they can autonomously handle tasks that stretch over longer time windows. Predictions of the labour-market impact diverge: conservative estimates suggest only a few percent of jobs may be automated in the near term, while more aggressive forecasts expect 10–20 percent of jobs, or up to half of all entry-level roles, to be automated in the next few years.
Picture both trends over time. We can be reasonably confident in how fertility decline will play out and how that will reduce the future workforce over 10, 20, 30 years. What is less clear is the overall impact on productivity: the number of workers multiplied by their output. Enhanced by AI, each worker may become far more productive. Whether that effect outweighs or falls short of the shrinking labour supply remains an open question.
Scenario 1: labour shortage ~ AI automation. The two forces cancel each other out. If AI lifts productivity enough to offset shrinking labour input, GDP per capita holds steady or even rises. This requires steady adoption of AI beyond the tech sector, plus significant investment in organisational change and skills.
Scenario 2: labour shortage > AI automation. Where are the robots when we need them? AI adoption stalls as models plateau and/or organisational barriers persist. At the same time, demography keeps tightening. Economic growth slows, wages rise sharply in shortage sectors, and fiscal stress mounts as pensions and healthcare lean on an ever-smaller workforce.
Scenario 3: labour shortage < AI automation. Fewer humans are still too many. AI capabilities keep advancing, automating complex tasks faster than the labour force shrinks. Millions of jobs disappear, first in routine white-collar roles, then in physical sectors as robotics catch up. Managing the transition from human to AI labour becomes the central challenge.
Reality will look be more complex still. A country could simultaneously be in Scenario 1 for highly specialist healthcare, Scenario 2 for elderly care delivery and Scenario 3 for back-office finance.
Importantly, the jobs societies need more of in the near to mid-term are rarely the ones AI threatens first. In the US, the Census Bureau’s fastest-growing occupations include wind-turbine technicians, solar PV installers, nurses, physician assistants, and personal-care aides. Meanwhile, Microsoft and the WEF list the most at-risk jobs as routine, codified roles: bank tellers, postal clerks, cashiers, ticket clerks, translators, and telephone operators. Projections estimate that Europe will be short of tens of millions of workers by 2050, but even with AI automating a significant amount of jobs, we could still see…
…Scenario 4: labour shortage & AI automation. The double whammy. There continues to be a significant shortage of workers in areas that are difficult to automate (like elderly care), while AI is automating large numbers of jobs that humans actually still want to do (like clerk or customer service roles). The result: many people out of work, and many roles left unfilled at the same time. In theory, displaced workers could re-skill and move into areas where humans are desperately needed. In practice, this is not how things look today. Despite rising demand for care workers, technicians, and solar PV installers, these jobs are often filled predominantly by immigrants. Local citizens often do not want work that means night shifts, no home office, and chronic back pain and understandably so. Higher wages may help, but structural mismatches are likely to persist. The jobs people want ≠ the jobs society most needs ≠ the jobs AI will automate first. What an inconvenience.
But not all futures look so grim.
Scenario 5: labour shortage ∞ AI automation. The paradox of abundance. AI automation doesn’t just replace human work, it can also create more demand for it. As costs fall and productivity rises, entirely new markets may open up: personalised education, customised health services, new forms of entertainment, and other industries we can’t yet imagine. Instead of a simple “too few jobs,” or “too few workers,” the problem becomes “too few skilled workers savvy at utilising AI.” In the past, the IT boom eliminated typists but created software engineers, cybersecurity engineers, data analysts, etc. In this scenario, AI accelerates demand for human labour in areas where human taste, empathy, and physical presence remain irreplaceable.
Whatever projections I make today, one thing is certain: they will be wrong. The collision of shrinking populations and rising automation will never produce a single, universal outcome. It will unfold differently across sectors, countries and time horizons. What I do know is that neither curve can be ignored: demography will steadily erode the supply of human labour, and AI will steadily reshape the demand for it. Whether this brings balance or disruption depends not only on the pace of AI, but on how quickly we adapt - through policy, investment and skills - to tilt the transition in our favour. Observing the labour market over the coming decade will be nothing short of a thriller.
