Jobs / Office and administrative support
Can agents do office and administrative support work?
Scheduling, records, data entry, bookkeeping clerks, and other office work.
- US jobs
- 15,157,670 (9.7% of employment)
- Typical wage
- $50K a year (employment-weighted mean of occupation medians)
- Occupations
- 55 in O*NET 31.0
- Evidence
- 5 direct benchmarks, 4 also relevant, 7 context only
The answer today
Most tasksNo change since 10 Oct 2025.
Rule: Agents complete 60% or more of the tasks (or projects) on two direct benchmarks from independent institutions.
The evidence that sets this step
- OSWorld-Verified by HKU, Salesforce AI Research, Carnegie Mellon University, and University of Waterloo: 90.2% of tasks completed · Intelligence-Indeed Agent, 25 Jul 2026
- SpreadsheetBench by Renmin University of China: 83.1% of tasks completed · Qingqiu Agent, 23 Jun 2026
Next step: Nearly all tasks needs 90% or more on two direct benchmarks from independent institutions.
How much of the work is measured
25% of the work in this family is measured by direct benchmarks. Some of the work is measured (20% to 50%). On 10 Oct 2025: 10%.
Work = the O*NET 31.0 work activities of the family's occupations, weighted by how important each task is and by US employment. A benchmark measures an activity when its tasks make the agent do it and its grading checks it. Method.
| Work activity (O*NET) | Share of the work | Direct benchmark |
|---|---|---|
| Maintain operational records | 5.0% | Measured |
| Provide information to guests, clients, or customers | 5.0% | Not measured |
| Communicate with others about operational plans or activities | 4.6% | Measured |
| Process shipments or mail | 3.4% | Not measured |
| Operate office equipment | 3.3% | Not measured |
| Prepare documentation for contracts, applications, or permits | 3.3% | Not measured |
| Evaluate the quality or accuracy of data | 3.3% | Not measured |
| Perform administrative or clerical activities | 3.1% | Measured |
| Compile records, documentation, or other data | 3.0% | Measured |
| Schedule appointments | 3.0% | Measured |
Our reading
Agents complete most tasks on two direct benchmarks from independent institutions. On OSWorld-Verified, the top agent completes 90.19% of 361 desktop tasks, above the 72.36% human figure; the board does not name its model. On SpreadsheetBench, the best agent solves 83.11% of 912 spreadsheet questions. Longer desktop work is harder: 44.33% on OSWorld 2.0, whose tasks take skilled users a median of 1.6 hours. MCPMark Verified (96.06% of MCP tool tasks) is broad and sets no step. No direct benchmark measures providing information to clients or processing mail. A second direct benchmark at 90% or more would move the family to nearly all tasks.
Evidence: OSWorld-Verified, SpreadsheetBench, OSWorld 2.0, MCPMark Verified by Carnegie Mellon University, EVAL SYS, HKU, LobeHub, NUS TRAIL, Renmin University of China, Salesforce AI Research, Snorkel AI, and University of Waterloo. Written 24 Sep 2026.
Benchmarks for this work
Direct evidence
Work benchmarks mapped primary to this family. They set the step when they report how many tasks agents complete.
- 45.9%Claude Opus 4.6 · 16 Mar 2026Emerging
- OSWorld-Verifiedby HKU, Salesforce AI Research, Carnegie Mellon University and University of Waterloo90.2%Intelligence-Indeed Agent · 25 Jul 2026Saturated
Also relevant
These benchmarks cover part of this family's work, or have no result yet. They do not set its step.
- 54.5%arito · 13 Sep 2026Emerging
Context only
These benchmarks span many occupations, or they test how agents behave. They give context but do not set the step.
- 42.9%TTE-MatrixAgent + DeepSeek-V3.2 · 10 Nov 2025Emerging
Functions: Bookkeeping and month-end closeSpreadsheet workBusiness workflow automationEnterprise workflowsCRM operationsCustomer serviceWeb researchIncident responseProfessional deliverablesMCP tool useData engineering and SQLComputer useProcedure executionFinancial analysisFeature developmentLong-horizon autonomy
What we do not know
- Benchmarks measure tasks, not whole jobs. A job is a bundle of tasks, judgment, and relationships.
- Capability is not adoption. A task that agents can do in a benchmark may still be done by people at work.
- Most results come from single runs. Reliability over many runs is often unknown.
Occupations in this family
Sorted by US employment (BLS OEWS May 2025). O*NET sub-occupations share the employment figure of their SOC group. Work measured = the share of the occupation's O*NET work that direct benchmarks test.
How to cite
Credit the original work first: the benchmarks listed on this page, and their institutions.
Then, if you used this page:
Can Agents Work. "Can agents do office and administrative support work?." https://canagentswork.com/jobs/office-admin/ (accessed 2026-10-11). CC BY 4.0.@misc{caw-jobs-office-admin,
title = {{Can agents do office and administrative support work?}},
author = {{Can Agents Work}},
year = {2026},
howpublished = {\url{https://canagentswork.com/jobs/office-admin/}},
note = {Accessed 2026-10-11. CC BY 4.0}
}