Skip to content
STRATUSWORKFORCE SCAN
Scan

Stratus Workforce Scan

See which hours of the work AI could take on, and when.

Every US job, task by task: what AI could take on now and through 2030, what becomes a person checking its work, and what stays with people. For one job, or for every role in your company.

Scan your companyBrowse all 923 jobs

Free to look up. Data checked November 10, 2026. : the best models finish tasks of about 2.4 work days , projected from METR's April 2026 measurement and Epoch's November 2026 index.

Customer Service Representatives: a 40-hour week

  • AI could take on8 h
  • A person checks AI's work18 h
  • Desk work AI can't do yet0 h
  • Stays with people14 h

Customer Service Representatives, now: of a 40-hour week, AI could take on 8 hours, 18 hours become checking its work, 0 hours are desk work it can't do yet, and 14 hours stay with people.

The largest pieces: Investigate the validity of complaints (about 4 h, a person checks); Complete contract and account forms (about 4 h, a person checks); Check that fixes were made (about 3 h).

Estimates for a typical week of the job, if progress keeps its long-run pace, counting work the best models finish four times in five. Open the job for every task.

For companies

The same picture for every role in your company

An equipment and supplies distributor with 250 people: AI could take on about 43% of its paid hours now, about 225,100 hours a year and $14M of work, rising to 52% by the end of 2028.

Paste your own titles and headcounts and the scan does the same for your roles: the hours and their cost, role by role, at every date to 2030, and where to start. Your list never leaves your browser.

Scan your rolesOpen this company in the scan

The preview is free. The full report is $249 for one company, with the workbook and the slides (sample workbook, sample slides).

Where to start in that company

Document work AI could take on today that needs no license and is not severe if it goes wrong, largest first.

  1. 1

    Prepare proposals and presentations

    Sales Representatives, Wholesale and Manufacturing, Technical and Scientific Products, about 7,950 hours a year

  2. 2

    Estimate customer costs and savings

    Sales Representatives, Wholesale and Manufacturing, Technical and Scientific Products, about 6,680 hours a year

  3. 3

    Monitor the market and competitors

    Sales Representatives, Wholesale and Manufacturing, Except Technical and Scientific Products, about 4,970 hours a year

What happens to the freed hours is the company's call

Hours AI could take on are not hours saved. A company decides what to do with them: buy fewer hours, or put the same people on more and better work. Most do some of both. This map shows how the pieces connect, and which ones this site measures.

Each step reads: more of this brings more, or less, of the next. R marks a loop that reinforces itself, B one that holds back.

R1 Growth engine

  1. Hours freed↓ brings more
  2. More and better work↓ brings more
  3. Growth↓ brings more
  4. Volume of work↓ brings more
  5. Hours AI could take on↓ brings more
  6. Work handed to AI↓ brings more, back to Hours freed

Freed hours go to more and better work: customers, quality, new offers. Growth brings more work, more of it is work AI could take on, and that frees more hours. A reinforcing loop: it speeds itself up.

R2 Fewer hours to buy

  1. Hours freed↓ brings more
  2. Fewer hours to buy↓ brings more
  3. Lower cost of the work↓ brings more
  4. Setup↓ brings more
  5. Work handed to AI↓ brings more, back to Hours freed

Freed hours mean fewer hours to buy: slower hiring, less overtime and contract labor. The lower cost pays for the setup (systems, data, controls) that lets more work be handed over. Also reinforcing.

B1 Trust

  1. Work handed to AI↓ brings more
  2. Mistakes↓ brings less
  3. Trust in the output↓ brings more, back to Work handed to AI

More work handed over means more chances for a mistake where the stakes are high. A mistake costs trust, and less trust slows the handover. A balancing loop: it holds the others back. Evidence builds trust, which is why the site publishes worked examples of the model doing real tasks.

Checking takes time back

  1. Work handed to AI↓ brings more
  2. Checking work↓ brings less
  3. Hours freed

Where a mistake costs something, a person checks the model's work. That checking is new work, so it takes back part of the freed time. The site counts it task by task.

The outside driver

  1. AI capability↓ brings more
  2. Hours AI could take on

AI capability is the push from outside: METR measures the length of task the best models finish, and it keeps doubling. As it grows, more hours come within AI's reach and every loop turns faster.

  • Measured or tracked on this site
  • The company's decision
  • What follows
  • The outside driver, measured by METR

A model of how the parts connect, not a forecast of what any company will do.

Behind every number

Anyone can draw those arrows. The work is in measuring them: every task read one at a time with its reason written down, the model doing real tasks in the open, and every estimate dated and kept, so what was known when can be checked later.

One task, opened up

Investigate the validity of complaints

“Obtain and examine all relevant information to assess validity of complaints and to determine possible causes, such as extreme weather conditions that could increase utility bills.”

A task of Customer Service Representatives

What it is made of
  • Documents and data 60%
  • Business systems 40%
Why, in one sentence
“Pulling the records and weighing what they show.”
A mistake is
costly if wrong
How long it takes
About 25 min, from 13 min to 42 min
When AI could take it on
Already, for most of it
The model doing it, in the open
Claude Fable 5.2 attempted it on November 11, 2026 on constructed material and passed with named fixes against a rubric written beforehand. Read the attempt, unedited
  • Tasks read one at a time8,645 of 18,838

    Each with what it is made of, what a mistake costs and a sentence saying why.

  • Occupations reviewed402 of 923

    Largest first: 93% of US jobs so far.

  • Companies read from their own annual reports1,427 of 1,835

    Every US-listed company with 1,000 employees or more, the largest first.

  • Worked examples, graded and published20 of 500

    The tasks carrying the most desk hours in the economy, done by the model in the open.

A record that cannot be backdated. Every estimate is kept in a dated snapshot (9 since September 29, 2026) and scored against what METR measures later. The record

Built on O*NET's task lists, BLS pay and employment, METR's measurements and Anthropic's Economic Index. How it works

Who it is for

Your own job, task by task

Give your title and paste your job description: what AI could take on and when, what stays with you, and what to learn. The headline is free; the full personal report is $19. See a sample.

The preview is free. A full report is $249; Pro is $299 a month. Pricing

Every figure here is an estimate with a stated method. “AI could take on” is the plain name for what the rest of the site calls : a model could finish the work at that reliability if the work were set up for it. It is not a forecast of jobs lost or of savings.