How is working life affected by AI?
Many fear high unemployment when AI arrives and takes the jobs, while others think people will find new work. The index gives running answers to these questions and is updated every month, so the conclusions are as fresh as possible. It compares employment in the occupations where language models can do most of the work with the occupations where they can do least. For more detail, read here.
What we mean by AI exposure, and how occupations are split into five groups
Every occupation is a set of tasks. Eloundou et al. (2024) rated, task by task, whether a language model can halve the time a task takes. The share of such tasks is the occupation's AI exposure, a score from 0 to 1. The score says what AI can do in the job, not whether the job disappears. Many of the tasks can be strengthened rather than replaced.
Highly exposed are occupations where much of the work is text, numbers and code: software developers score 0.87, credit and loans officers 0.61, general office clerks 0.60 and accountants 0.57. Low exposure is found where the work is done by hand and on site: electricians 0.13, health care assistants 0.11, plumbers 0.06 and cleaners 0.03.
We rank 397 occupations by the score and split them into five equal groups: least exposed, low, medium, high and most exposed. The index compares employment in the most exposed fifth with the least exposed. Look up your own occupation under Occupations.
All ages, 21–60
The youngest, 21–30
The lines are drawn month by month. The dashed lines are the two events: ChatGPT in November 2022 and agentic AI, with Claude Code, in February 2025. The index is the relative growth between the two lines from the event to the latest month.
Seasonally adjusted private-sector employment, index 100 in November 2022. Occupations are split into five equal groups by AI exposure (Eloundou et al. 2024). The figures show the most and the least exposed fifth. The index numbers use a three-month moving average, as on the index page, which also has the uncertainty, more measures and the method. Source: the A-ordningen via microdata.no.