A Practical Way to Read Your Own Job's AI Exposure

Most of the anxiety around AI and work attaches to job titles: is 'translator' on the list, is 'accountant' safe. But the more useful unit, and the one recent research increasingly uses, is not the title — it is the task. A job is a bundle of tasks, and AI does not arrive for all of them at once.

Titles hide the variation; tasks reveal it

Anthropic's Economic Index, which studies how people actually use its AI assistant Claude at work, has framed this well by looking at tasks rather than occupations. In its January 2026 report, pooling data across its reports, about 49% of occupations had Claude used for at least a quarter of their tasks, up from 36% in the first report (January 2025 data). But the pattern within a job matters more than the headline. The report adds that coverage alone can mislead: once tasks are weighted by how successfully Claude completes them, a different picture emerges of which jobs are most affected. In our reading, a role can look lightly exposed by task count yet be heavily exposed in practice, if the AI happens to be good at the tasks that take the most time.

The same research distinguishes between AI that automates a task (does it instead of you) and AI that augments it (does it alongside you). On Claude.ai the split has run close to even: in the latest sample augmentation accounted for 52% of conversations and automation for 45%, after automation briefly led in an August 2025 sample. For your own planning, that distinction is most of the game: an augmented task changes shape, while an automated one can disappear.

A short audit you can do on one page

Here is the exercise we would suggest, and it takes about fifteen minutes. Write down the tasks that actually fill your week — not your job description, but where your hours really go. Against each one, mark two things: how routine and repeatable it is (does it follow a stable pattern, or need fresh judgement each time), and how much of your time it eats. The tasks that are both highly routine and time-consuming are your most exposed; the tasks that need judgement, accountability, or a human relationship are, for now, your most durable.

Your job title tells you very little. The mix of tasks inside it — weighted by where your hours actually go — tells you almost everything.

What comes out of that page is not a score. It is a shape: which parts of your work sit on the exposed side, and which sit on the durable side. That shape is more honest than any single number attached to your title, because it is yours.

Turning the map into a move

Once you can see the shape, the response follows from it. The routine, exposed tasks are the ones to start handing to tools deliberately, so the time they free up can move toward the durable side — the exceptions, the judgement calls, the relationships. That is the same direction the labour market seems to be pushing anyway; doing it on purpose is simply getting there first.

If it helps to start from an outside estimate rather than a blank page, our risk checker walks through a similar task-based set of questions, and our comparison tool lets you hold two roles side by side. But treat both as prompts for your own audit — the version you write about your own week will always be the accurate one.