United States & United Kingdom
Will AI take your job? That’s the wrong question.
Look up your job. See what AI can really do. Know what to do next.
It is not coming for your job. It is coming for parts of your day: some of them the parts you are quietly bored of, some of them the parts you are paid for.
So we score the tasks, not the job title. Every documented task in an occupation, rated against what today’s AI can actually do, so you can see which parts of your week are shifting, which are changing shape, and which are still unmistakably yours.
1,242 occupations scored so far: 830 in the US and 412 UK SOC 2020 unit groups, from 53,309 scored task statements.
A few jobs, across the range
Deliberately a spread, not a ranking. Some jobs are barely touched; some have had a third of their week rewritten. Both facts are useful and neither is a headline about anyone’s future.
US
Proofreaders and Copy Markers
very high exposure
- shifting to AI
- 81%
- changing shape
- 15%
- staying human
- 4%
Share of task weight across 11 scored tasks. Exposure describes how the work changes. It is not a prediction that anyone loses a job.
US
Architecture Teachers, Postsecondary
low exposure
- shifting to AI
- 21%
- changing shape
- 23%
- staying human
- 57%
Share of task weight across 22 scored tasks. Exposure describes how the work changes. It is not a prediction that anyone loses a job.
US
Hosts and Hostesses, Restaurant, Lounge, and Coffee Shop
minimal exposure
- shifting to AI
- 8%
- changing shape
- 7%
- staying human
- 85%
Share of task weight across 20 scored tasks. Exposure describes how the work changes. It is not a prediction that anyone loses a job.
UK
Street cleaners
minimal exposure
- shifting to AI
- 0%
- changing shape
- 0%
- staying human
- 100%
Share of task weight across 6 scored tasks. Exposure describes how the work changes. It is not a prediction that anyone loses a job.
Two countries. Two real datasets. No translation tricks.
A UK bookkeeper does the VAT return. A US bookkeeper files sales tax. Pretending those are the same job, and quietly showing a British reader a dollar figure, is the fastest way to lose them, and rightly so.
So the two markets are built separately. US pages score O*NET task statements and carry BLS pay and employment. UK pages score UK-native task statements against ONS SOC 2020 unit groups, with ASHE pay and Nomis employment. The method, the rubric and the model are identical; only the facts and the vocabulary change.
Where a job maps cleanly across the Atlantic you can switch between them. Where the mapping is rough, the page says so and shows you the quality grade instead of pretending.
- United States
- 830occupations scored, of a target 785 SOC occupations.Tasks from O*NET. Pay and employment from BLS OEWS.
- United Kingdom
- 412SOC 2020 unit groups scored, of all 412.Tasks from the UK Standard Skills Classification. Pay from ONS ASHE. Employment from Nomis.
How it works
Three steps, all of them checkable
01
We take the job apart
Occupations are not units of work; tasks are. We start from the official task statements published for every occupation (the actual sentences describing the actual work), so the thing being scored is “reconcile supplier statements”, not “bookkeeper”.
02
We score each task against what AI can do today
Every task is rated on five dimensions: can a model produce the output, does the work need a body in a room, does it require a legally accountable human, and so on. The model never writes the score: it writes the five ratings, and a published formula turns them into a number you can recompute by hand.
03
We publish the whole apparatus
The prompt, the rubric, the formula, the dimension ratings, the source list with fetch dates and content hashes, and the full dataset as CSV. Every release keeps its own permanent address, so a figure you cite in March is still there, unchanged, in November.
What the number is, and what it isn’t
Exposure means transformation, not termination
A high score means today’s AI could already do much of the work inside a job’s tasks. It is a statement about the tasks. It is not a statement about whether the job survives, whether demand grows, or whether you personally are all right.
We will not soften that in either direction. Where the exposed share is large we say so plainly, because being told a comfortable half-truth at 11pm is not a kindness. And we will not tell you that you are going to lose your job, because we cannot know that, and claiming it would be an invented certainty.
What this method deliberately ignores
- Whether your employer actually adopts any of it.
- Whether customers accept being served by a machine.
- Whether the law permits it in your field.
- Whether doing the routine parts faster creates more demand for the human parts, which has happened before, more than once.
Those forces are real and they matter enormously. We do not model them, so we do not pretend to. That boundary is written into the method page, not buried in a footnote.
Where to go next
Two honest options, and no deadline on either
Free, and complete
Every score, every task ledger, every piece of advice and the full dataset are free and always will be. Nothing on this site sits behind an email address. Start with your own job, then read how it was scored.
With other people, if you’d rather not do it alone
This site is built by Collab365, which runs Spaces: paid, guided communities where people work through real problems with AI alongside others doing the same job. It is genuinely useful to some people and genuinely unnecessary for others, and you will get identical data either way.
Figures on this page come from release 2026-q4.1. Release 2026-q4.1, read from the published data store.