Microsoft’s research division has published one of its most detailed examinations yet of generative AI’s labour-market impact, analysing 200,000 anonymised conversations with its Copilot chatbot to measure how closely large-language-model capabilities align with specific job tasks. The paper assigns an “AI applicability” score to 80 occupations and finds that knowledge-heavy roles face the highest overlap. Interpreters and translators top the list with a score of 0.49, followed by historians, passenger attendants, sales representatives and writers. Other highly exposed roles include customer-service representatives, CNC-tool programmers, telephone operators, ticket agents and broadcast announcers — jobs that collectively account for millions of positions in the United States. At the opposite end of the spectrum are 40 largely manual or in-person occupations such as dredge operators, bridge-and-lock tenders, nursing assistants, phlebotomists and hazardous-material removal workers, where current generative models show little capability to automate core tasks. Senior researcher Kiran Tomlinson cautioned that a high overlap indicates where AI could change how work is done rather than fully replace human labour. Still, the findings add quantitative backing to mounting evidence that generative AI is poised to reshape white-collar employment while leaving many physically demanding roles relatively insulated for now.