Worked examples
The Work Bench
The numbers on this site rest on how long a task AI models can finish, measured by METR on software tasks. The Work Bench looks at the economy's own tasks instead: for a common task from the task file, a realistic brief and material, a rubric written and timestamped before any attempt, one attempt by a named model on a named date, unedited, and the grade against the rubric. Anyone can read the output and judge it for themselves.
20 examples across 13 occupations so far, attempted by Claude Fable 5.1 (October 1, 2026): 19 passed, 1 passed with named fixes, 0 failed. Tasks are taken in order of the desk hours they carry across the US, among tasks that are mostly documents work: the 500 that carry the most first, shorter ones first among them, so the bench grows from the work that matters most.
The Work Bench number, so far
The 500 tasks that carry the most documents work across the US, among the jobs reviewed task by task so far, hold about 45 billion desk hours a year. 20 of them have an attempt, holding 8.9% of those hours: 8.6% passed, 0.3% passed with named fixes and 0.0% failed. The number is the share of the 500's hours whose latest attempt passed, published with its date once all 500 have one; until then it is a count of progress, not a result. Desk hours here are US employment times 2,080 hours times each task's share of working time times its documents share; the set moves a little as the review reaches more jobs.
Every example is one attempt at one task on one date, with material constructed for it: it is not a statement that a job, or the task in a given workplace, can be done by AI. The grade is the model's own against the written rubric; the rubric, the attempt and the grade were made by the same model in one session, in that order, and the timestamps show it. A person's check of a sample of grades comes later and will be published beside them. When a new model is available the same briefs are run again, and earlier attempts stay on each page.
Customer Service Representatives
- Investigate the validity of complaints · passes
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. Typical length 25 minutes; 352,489,560 desk hours a year across the US.
- Review policy terms for coverage · passes
Review insurance policy terms to determine whether a particular loss is covered by insurance. Typical length 17 minutes; 316,336,784 desk hours a year across the US.
- Recommend service improvements · passes
Recommend improvements in products, packaging, shipping, service, or billing methods and procedures to prevent future problems. Typical length 33 minutes; 119,299,839 desk hours a year across the US.
Office Clerks, General
- Compute, record and proofread data · passes
Compute, record, and proofread data and other information, such as records or reports. Typical length 38 minutes; 308,701,198 desk hours a year across the US.
- Type and edit documents · passes
Type, format, proofread, and edit correspondence and other documents, from notes or dictating machines, using computers or typewriters. Typical length 25 minutes; 293,986,492 desk hours a year across the US.
- Look up records to answer requests · passes
Review files, records, and other documents to obtain information to respond to requests. Typical length 17 minutes; 122,152,567 desk hours a year across the US.
Retail Salespersons
- Keep up with promotions and policies · passes
Maintain knowledge of current sales and promotions, policies regarding payment and exchanges, and security practices. Typical length 25 minutes; 472,305,255 desk hours a year across the US.
Stockers and Order Fillers
- Decide storage methods and locations · passes
Determine proper storage methods, identification, and stock location, based on turnover, environmental factors, and physical capabilities of facilities. Typical length 29 minutes; 210,073,736 desk hours a year across the US.
- Recommend disposal of excess stock · passes
Recommend disposal of excess, defective, or obsolete stock. Typical length 25 minutes; 129,115,185 desk hours a year across the US.
Secretaries and Administrative Assistants, Except Legal, Medical, and Executive
- Use word processing and databases · passes
Use computers for various applications, such as database management or word processing. Typical length 38 minutes; 170,459,165 desk hours a year across the US.
- Write notes, letters and reports · passes
Compose, type, and distribute meeting notes, routine correspondence, or reports, such as presentations or expense, statistical, or monthly reports. Typical length 50 minutes; 122,266,243 desk hours a year across the US.
Registered Nurses
- Order and interpret diagnostic tests · passes with named fixes
Order, interpret, and evaluate diagnostic tests to identify and assess patient's condition. Typical length 38 minutes; 140,982,992 desk hours a year across the US.
- Modify treatment plans · passes
Modify patient treatment plans as indicated by patients' responses and conditions. Typical length 25 minutes; 128,582,394 desk hours a year across the US.
Security Guards
- Write daily activity reports · passes
Write reports of daily activities and irregularities, such as equipment or property damage, theft, presence of unauthorized persons, or unusual occurrences. Typical length 25 minutes; 242,748,328 desk hours a year across the US.
Sales Representatives of Services, Except Advertising, Insurance, Financial Services, and Travel
- Monitor the market and competitors · passes
Monitor market conditions, innovations, and competitors' services, prices, and sales. Typical length 50 minutes; 191,391,809 desk hours a year across the US.
Medical Secretaries and Administrative Assistants
- Prepare reports, letters and records · passes
Operate office equipment, such as voice mail messaging systems, and use word processing, spreadsheet, or other software applications to prepare reports, invoices, financial statements, letters, case histories, or medical records. Typical length 25 minutes; 150,105,167 desk hours a year across the US.
Bookkeeping, Accounting, and Auditing Clerks
- Compare printouts with journals · passes
Compare computer printouts to manually maintained journals to determine if they match. Typical length 38 minutes; 146,634,296 desk hours a year across the US.
Receptionists and Information Clerks
- Do proofreading, transcribing and data work · passes
Perform administrative support tasks, such as proofreading, transcribing handwritten information, or operating calculators or computers to work with pay records, invoices, balance sheets, or other documents. Typical length 25 minutes; 141,003,629 desk hours a year across the US.
Laborers and Freight, Stock, and Material Movers, Hand
- Read work orders or take instructions · passes
Read work orders or receive oral instructions to determine work assignments or material or equipment needs. Typical length 7 minutes; 139,024,274 desk hours a year across the US.
First-Line Supervisors of Food Preparation and Serving Workers
- Specify portions, sequences and station layouts · passes
Specify food portions and courses, production and time sequences, and workstation and equipment arrangements. Typical length 50 minutes; 122,942,470 desk hours a year across the US.
How an example is made and graded: the method.