A new open index covers 146 countries and 18,796 tasks: the share of delegable jobs rises from 22.5 per cent in poor countries to 48.7 per cent in rich countries.
Most existing measures estimate exposure to generative AI, and almost always for the United States alone. The Agent Delegability Index (ADI) asks another question: how much of the work can be entrusted to an AI agent, and to what extent? Two language models scored the 18,796 tasks in the O*NET 30.2 database according to a pre-recorded grid of levels of delegation and obstacles, before linking them to 433 jobs in the ISCO-08 nomenclature and weighting them by the actual employment of 146 countries according to the International Labour Organization.
A "fracture of the delegation"
As a result, by 2026, 35.2% of global working time would be at least partially delegable, compared to only 2.5% completely. The share of delegable jobs rose from 22.5% in low-income countries to 48.7% in high-income countries, i.e. 7.23 points more per logarithmic point of GDP per capita. The gap is due to the structure of economies: physical barriers dominate in the jobs that employ the most people in poor countries. Full delegation remains rare and peaks for tasks that require about twelve minutes for a human.
The hypothesis of a delay in adoption specific to middle-income countries is not confirmed: out of 101 countries, delegability and observed use of AI are correlated with 0.86. A blind human rating of 300 tasks is in line with the consensus of the models (kappa of 0.85). The index, the atlas, the prompts and an MCP server are published in open access.
Read the full article on the original source
Go to the source →