If your job mostly involves talking, writing, or explaining things to other people, Microsoft’s own data suggests you should pay closer attention to how AI is already being used at work than someone whose job involves fixing, building, or physically caring for something.
Researchers at Microsoft, the company behind Copilot, analyzed roughly 200,000 anonymized conversations between users and the AI assistant, then matched that usage against measures of task success and scope of impact to build what they call an AI applicability score for 80 different occupations. The resulting rankings split cleanly into two lists: jobs where Copilot use is already heavy, and jobs where it’s barely showing up at all.
The most AI-exposed occupations cluster almost entirely around communication and information work. Interpreters and translators top the list, followed by historians, sales representatives, writers and authors, and customer service representatives. Journalists, editors, technical writers, proofreaders, public relations specialists, market research analysts, and data scientists all land in the top 40 as well. The researchers describe the pattern directly: the highest AI applicability scores show up in “knowledge work occupation groups such as computer and mathematical, and office and administrative support, as well as occupations such as sales whose work activities involve providing and communicating information.”
The least-exposed list looks almost like a mirror image. Phlebotomists, nursing assistants, hazardous materials removal workers, embalmers, and oral and maxillofacial surgeons show minimal AI usage in Microsoft’s data, alongside roofers, dishwashers, tire repairers, cement masons, and dredge operators. The common thread across nearly all 40 jobs on this side of the list is some combination of hands-on manual work, operating physical machinery, and direct in-person interaction, exactly the kind of work a chat-based AI tool has no way to touch yet.
Microsoft’s researchers are notably careful not to translate that usage data into a prediction about actual job losses, and they say so explicitly. “It is tempting to conclude that occupations with high overlap will experience job loss,” they write. “This would be a mistake, as our data do not include the downstream business impacts of new technology, which are very hard to predict.” They point out that even within heavily-AI-using professions, workers show up in the data using Copilot in very different ways: some ask the tool to complete entire tasks outright, while others use it more like a research assistant, asking how best to do the work themselves rather than handing it over entirely.
The researchers also resist the idea that heavy AI use in an occupation automatically translates into fewer jobs in that occupation. They lay out the underlying ambiguity plainly, using software developers as their example: if AI makes developers 50% more productive, a company could respond by hiring more developers to take on more ambitious projects with that extra output, or by hiring fewer developers to get the same amount of work done. “Our data is only about AI usage and we have no data on the downstream impacts of that usage, so we only weigh in on the automation versus augmentation question by separately measuring the tasks that AI performs and assists,” they write.
There’s an important caveat buried in the study’s own methodology, one that cuts against reading too much long-term comfort into the “least exposed” list. Microsoft’s researchers point out their entire analysis is built on usage of large language models specifically, the technology underlying tools like Copilot and ChatGPT. “Other applications of AI could certainly affect occupations involving operating and monitoring machinery, such as truck driving,” they note. In other words, a roofer, a pile driver operator, or a dredge operator showing up near the bottom of today’s exposure list isn’t necessarily protected from AI long-term. They’re protected from this specific generation of chat-based AI tools, which simply weren’t built with hands and machinery in mind. More specialized AI, paired with robotics or industry-specific tools, could reach those jobs eventually even if general-purpose chatbots haven’t yet.
The study lands in the middle of a genuinely unsettled debate among the people building this technology. Anthropic CEO Dario Amodei has warned AI could eliminate as much as half of all entry-level white-collar jobs. Serial entrepreneur Mark Cuban has argued the opposite, that AI will end up creating more jobs than it destroys by absorbing repetitive work humans don’t want to do anyway. Microsoft’s own study, notably, refuses to take a side in that argument, offering instead a snapshot of where AI use is concentrated today rather than a forecast of where employment is headed tomorrow, an important distinction for anyone reading their own job title into either list.