A Singulariki data story

Where the world's work sits

Scroll the global GenAI exposure gradient: 427 international occupations from the human ground of physical work to the hot end of clerical and text work — and how it moved from 2023 to 2025.

2025 — task exposure (ILO / ISCO-08)

Each dot is an ISCO-08 occupation. Colour = exposure. International data; task overlap, not automation.

This is the world's work, measured task by task.

The International Labour Organization scored thousands of tasks across 427 international (ISCO-08) occupations for overlap with generative AI. Here they are, cool (least overlap) on the left to hot (most) on the right.

The cool end stays human.

Bricklayers, vehicle cleaners, farm labourers, carers — physical, in-person, bodily work. Today's AI overlaps their tasks the least. This is the human ground.

The hot end is clerical, data and text work.

Data-entry clerks, typists, accounting and finance clerks. Work made of structured language and numbers — exactly what large language models handle. The most exposed work on Earth is paperwork.

And it is moving. This was 2023.

Two years ago the gradient sat a little cooler. Watch what happens when the same occupations are re-scored in 2025.

2025: the middle heated up.

Software, web and database work moved the most — as tooling caught up, models began assisting tasks that were out of reach in 2023. The shift is concentrated in technical, screen-based work.

Most of the world's work is still on the cool side.

Exposure is task overlap, not a verdict. Low overlap is not “safe forever,” and high overlap is not automation — but as a map of where AI touches work today, the gradient is unambiguous: it runs from hands to keyboards.