Artificial intelligence is widely expected to transform the economy, and companies are investing vast sums in the hope of higher productivity and profits. Yet Greg Ip, an economics commentator for The Wall Street Journal, highlights a puzzling feature of the current boom: strong investment and economic activity have not been matched by equally strong hiring. His argument is not simply that AI will destroy jobs, but that it may weaken the traditional link between economic growth and employment. I find this concern persuasive, but the evidence does not yet show that the link has been broken. The more pressing question is whether the gains from AI-driven growth will be broadly shared.
Ip's argument matters because it challenges a familiar chain of reasoning. Traditionally, technology that raises productivity lowers costs and therefore prices. Lower prices increase consumers' purchasing power, which boosts demand, encourages firms to expand and eventually creates jobs. Generative AI may weaken this chain because it performs tasks once reserved for educated workers, such as drafting documents, writing code and analysing data. Imagine a firm that needs 1,000 employees to produce $100 million of output. If AI allows it to produce $150 million with 900 employees, its productivity has risen sharply, yet it needs 100 fewer workers. Klarna's 2024 claim that its AI assistant was handling work equivalent to roughly 700 customer-service agents is one real-world example of AI substituting for labour. If growth increasingly comes from capital rather than labour, that would explain why healthy headline figures can coexist with widespread anxiety about jobs.
However, this example is incomplete because it stops at the firm's gates. If the firm cuts prices, or its owners spend their higher profits, demand for other goods and services rises and jobs may appear elsewhere. History suggests this is what usually happens. Personal computers eliminated many clerical tasks but created demand for software developers and IT specialists. The internet disrupted traditional retail and media but gave rise to e-commerce, digital advertising and cloud computing. These effects take years to emerge, as firms redesign processes and workers acquire new skills. Today's weak hiring may therefore reflect a transitional stage rather than a permanent change. The real test is not whether AI destroys jobs, since technology clearly can, but whether it creates new opportunities quickly enough and on a large enough scale to offset those it displaces.
There are also reasons to doubt that AI is the main cause of weak hiring. Much of the AI boom visible in growth figures consists of investment in chips, data centres and power infrastructure, spending that is inherently capital-intensive. Such investment would lift output without creating many jobs even if AI replaced no workers at all. Firms also make hiring decisions in response to interest rates, consumer demand and uncertainty, so if graduates struggle to find work, high borrowing costs or cautious sentiment may be as responsible as automation. Even Klarna later began recruiting human agents again amid concerns about service quality, suggesting that substitution may not be as straightforward as early claims implied. To establish Ip's thesis, researchers would need to separate AI's effects from these wider forces. A correlation between AI investment and weak hiring is not, by itself, proof of causation.
Nevertheless, even if AI eventually creates as many jobs as it displaces, a harder problem remains: who receives the gains? AI need not simply replace workers. A study of customer-support agents by Erik Brynjolfsson, Danielle Li and Lindsey Raymond found that an AI assistant raised productivity by around 14 per cent, with the largest gains among the least experienced. Whether such gains become higher wages, however, depends on who holds bargaining power. If firms can produce more with fewer people, shareholders and workers whose abilities complement AI may capture most of the rewards, while those whose tasks are easily automated face lower pay or fewer openings. Transitions are also rarely smooth. An administrative worker whose tasks are automated cannot simply become a software engineer after a short course, and new jobs may demand different qualifications, be in different places or pay less. The impact of AI should therefore be judged not only by whether total employment recovers, but also by whether displaced workers have realistic routes into the new economy.
This suggests that the familiar question, "Will AI take our jobs?", is too narrow. A better question is how societies should distribute the benefits of growth that may require less labour. Governments can influence whether these gains are broadly shared through education, retraining and social protection, while firms can choose whether to deploy AI primarily to replace workers or to augment them. These choices matter because the same technology can produce very different labour-market outcomes depending on how it is adopted. Ip is right, then, to challenge the assumption that growth automatically brings broad-based employment, but it is too early to conclude that AI has permanently broken that relationship. What is already clear is that GDP growth, job creation and rising living standards are three different outcomes. Whether AI delivers all three will depend not only on the technology itself, but also on how societies, businesses and governments respond to it.