Skip to content

Which parts of your job are people already doing with AI?

Type your job and see its tasks, as the US Department of Labor lists them, placed on a handover scale from this year's data. Every task says where its place comes from.

  • One job title
  • About a minute
  • Nothing you type or tick is sent

Built for the essay The Safe Parts of Your Job Are the First to Go

Check my jobHow it works
Every task in the US job database…
…

of all the tasks in all the jobs show up in AI use.

    April and May 2026; the bar shows people's own chats. Few tasks are handed to the AI outright in chats.
    01 · Your job

    Type the job you do

    Pick the closest match from the list. Many UK titles work too (solicitor, GP, head teacher, lorry driver). If yours does not, try the nearest US title or build your own list from tasks.

    Try

    No good match? .

    Every job is a bundle of tasks, and some show up in Claude use far more than others.

    The essays on which parts of your work keep their value as AI gets cheaper come with sources shown. Subscribing is free.

    Subscribe opens Substack in a new tab with your address. Nothing you entered in the tool goes with it.

    The next step is your own week.

    This check reads your job as the Department of Labor writes it. The Working Week Audit reads the week you actually work: four questions per task, a saved history and a re-check in 90 days. It is for paid subscribers, and the essay below opens it.

    The Safe Parts of Your Job
    02 · How it works

    Two sources, read as they are

    One says what a job is made of: the US Department of Labor's O*NET. The other says what was done with Claude in April and May 2026: Anthropic's Economic Index, in two parts, people's own chats and calls through Anthropic's API. Each task gets its place from that data alone. There is no overall score and no forecast.

    1. Seen in useThe task shows up in Claude.ai conversations in April or May 2026 (Anthropic Economic Index, task level). "Seen rarely" means Anthropic published it but its share rounds to 0.00%.
    2. Handed over or worked through togetherFor tasks with enough conversations, Anthropic splits them: handed over means the person asked the AI to do it; worked through together means they did it with the AI, step by step. A task is "mostly handed over" at 55% or more, "mostly worked through together" at 45% or less.
    3. No clear leanIn use with a measurable share, but either too few conversations to split, or between 45% and 55% handed over.
    4. Not seen in chatsAnthropic published no use of the task in people's chats in either month. Use below its publishing threshold, or in other AI tools, does not show up here.
    5. In API useA second, separate signal: the task shows up in calls to Claude through Anthropic's own API (Claude Code excluded). Anthropic's file does not say who makes those calls. It never moves a task on the scale; it is marked on the task and counted on its own row. Where there is enough API traffic to tell (450 tasks), Anthropic classes 98% of them as mostly handed over: its "automation" category.
    Part of Claude, no other AIUse comes from two routes into Claude: people's own chats (Free, Pro and Max plans) and Anthropic's own API. Claude Code, Team and Enterprise plans, and Claude reached through other companies' clouds are not in these files; nor is ChatGPT, Gemini or Copilot. "Not seen" means Anthropic published no use for that task, which can also mean use below its publishing threshold; "seen" is a floor.
    Shares are roundedAnthropic publishes shares to two decimals and spreads them over about 2,600 tasks a month. Most tasks hold a tiny share: about four in five read 0.02% or less.
    The split needs volume"Handed over" against "worked through together" is shown only for tasks that reach a 0.03% share in at least one month. Below that, the median task moves about four points from one month to the next, and one in ten moves 14 or more.
    US job listsO*NET describes US jobs. UK and newer titles are mapped by hand to the nearest US job, and the match says when it is only the nearest.
    Two months of dataThe scale reads April and May 2026, the months in Anthropic's latest release. A later release can move a task; the colophon names the months read.
    No forecastThe check shows where use stood in April and May 2026. It says nothing about what happens to any one person's job.
    The rules in full, the files and the licences

    Use. Anthropic Economic Index, release of 26 June 2026, two files: Claude chat and Cowork conversations on the Free, Pro and Max plans, and first-party API traffic (excluding Claude Code), each classified to O*NET tasks, global, months April and May 2026. A task counts as seen if it was published in either month. Its share is the mean of the months it was published in. "Handed over" is Anthropic's automation share (the person delegates the task); "worked through together" is its augmentation share; the two add to 100%, to rounding. "For work" is the share of those conversations classed as work rather than personal or coursework. The time estimates are Anthropic's own: how long the task would take a person without AI, and with it. All of these are shown only for tasks that reach a 0.03% share in at least one month, and averaged over the months that reach it. A task missing from the release may still be done with AI: Anthropic publishes a task only above its own thresholds.

    Jobs and titles. O*NET 31.0: 923 occupations with task statements, their lay titles and reported titles, plus 156 hand-made aliases for UK and newer titles, published with this page. Where one title fits several jobs, up to twelve are listed, closest first; add a word to narrow.

    What the page downloads. The title list once, then one file of jobs for the field you pick (for example Computer and Mathematical), and the full task list only if you build your own list. Our host can see which file was fetched, as with any page; it never sees the title you typed or the tasks you tick. No cookies, no analytics script on this page.

    Files. manifest.json (sources, hashes, counts) · occupations.json · titles.json (with the aliases) · tasks.json.

    Data as of: Anthropic Economic Index release of 26 Jun 2026 (Claude.ai use and first-party API traffic, Apr and May 2026) · O*NET 31.0.

    This page includes information from the O*NET 31.0 Database by the U.S. Department of Labor, Employment and Training Administration (USDOL/ETA). Used under the CC BY 4.0 license. O*NET® is a trademark of USDOL/ETA. The Durability Curve has modified all or some of this information: tasks are grouped by occupation and joined to the use data below, and titles are added by hand (the aliases in titles.json). USDOL/ETA has not approved, endorsed, or tested these modifications.

    Use data: Anthropic Economic Index, CC BY.