Ghost GDP or unexpected jobs: Making sense of what lies in the AI-led future

IT-led growth and global outsourcing.How far have AI’s capabilities advanced to significantly affect employability across professions?A recent study by Anthropic, based on real-world usage of its Claude model, maps the share of tasks within occupations that large-language models (LLMs) can currently perform. Exposure is highest in digital and information-processing roles.

Computer programmers top the list, with about 75% of tasks covered, followed by customer service representatives at around 70% and data entry keyers at about 67%.Other highly exposed roles include medical record specialists, market research analysts, and sales representatives. Even technical roles, such as financial analysts and software testers, show exposure above 50%.However, even in the most affected occupations, this exposure has not yet translated into job losses, the study finds based on survey data from the US, although there is tentative evidence that hiring has slowed slightly for younger workers aged 22 to 25.The research underscores that most occupations consist of a bundle of tasks, many of which require judgment, context, accountability, and human interaction, areas where AI still struggles.

Even in highly exposed roles, AI is more often used to assist rather than replace workers.The limited translation of task exposure into job losses, as Anthropic finds, reflects what economists call the “lump of labour fallacy”, the mistaken belief that there is a fixed amount of work in an economy and that automation simply replaces workers one-for-one.In reality, labour markets evolve as technology raises productivity and creates new demand for goods, services, and entirely new occupations. The World Economic Forum estimates that while about 92