The government mapped 831 occupations against AI. Exposure is not the same as replacement

The largest projected loss sits inside everyday office work

BLS projects that office and administrative support employment will decline 4.0 percent from 2025 to 2035. That works out to 752,100 fewer jobs, the largest projected loss among the 22 major occupational groups.

The agency points to automation, including tools powered by AI, as one force likely to reduce demand for some of that work. It also projects a 1.4 percent decline in sales occupations and a 0.4 percent decline in production occupations as digital sales systems and automated machinery spread.

Those are group-level estimates across a decade. They do not say when a particular employer will automate a task, which workers will be affected, or whether a job that changes will disappear. BLS advises readers to focus on the direction and relative size of projected changes rather than treating each estimate as a precise outcome.

The fastest growth is arriving from several directions

Healthcare support is projected to grow 13.3 percent, faster than every other major occupational group. Healthcare practitioners and technical occupations follow at 8.0 percent. Together, those two groups are expected to account for almost one-third of the jobs added through 2035.

At the level of individual occupations, nurse practitioners lead the forecast at 41.0 percent growth. Solar photovoltaic installers are projected to grow 36.5 percent, data scientists 34.6 percent, wind turbine service technicians 29.5 percent, and computer and information research scientists 21.8 percent.

The percentages need context. Solar photovoltaic installers and wind turbine service technicians begin from small employment bases, so their fast growth is projected to add fewer than 15,000 positions combined. By comparison, the services for the elderly and persons with disabilities industry is projected to add 625,400 jobs.

AI demand creates jobs outside AI job titles

The fastest-growing major industry sector in the forecast is utilities, at 9.8 percent. BLS expects the sector to add only 58,800 jobs because it is relatively small, but the reason for the growth reaches far beyond a household power bill. The agency says demand from data centers and other AI infrastructure is helping push electricity needs higher.

Computing infrastructure, data processing, web hosting, and related services are projected to grow 25.1 percent and add 120,400 jobs. BLS projects growth in other electrical equipment and component manufacturing partly because of demand for batteries used in energy storage and electric vehicles. It separately says demand for fiber-optic cables is rising for AI and telecommunications infrastructure.

That produces a less familiar picture of the AI labor market. Some of its effects may appear in a data-science role. Others may show up in power generation, construction, equipment maintenance, manufacturing, or the care economy that is growing for reasons largely unrelated to software.

Exposure measures contact with AI, not a verdict on the worker

For the new exposure map, BLS combined five research sources. Three estimate how closely current AI capabilities match occupational abilities or tasks. Two use observed interactions from Claude and Microsoft Copilot that researchers mapped to occupational work.

The result places each occupation into one of four relative categories: low, moderate, high, or very high exposure. A very-high designation means an occupation has more tasks that AI could assist with or complete and more evidence of AI performing related work than other occupations in the dataset.

It does not mean the job is likely to vanish. BLS says exposure is not an estimate of job loss, adoption, productivity, wages, or worker replacement. The observed-use measures also do not directly establish that employees in a particular occupation used AI while doing their jobs.

The map is current, but the technology will not stand still

The agency built the categories from research that captures different moments in AI development. Its theoretical sources describe capabilities available no later than mid-2023. The observed sources use more recent interactions but may reflect the behavior of early adopters of particular products.

BLS plans to update its employment projections every year. That matters because this first map is best read as a baseline: a careful record of where AI could touch work now, not a final answer about what 831 occupations will become.