AI and your job: Secretaries and administrative assistants
Keeps an office running: schedules, messages, documents, records and the many small tasks that let others do their work.
Will AI replace secretaries and administrative assistants?
No published measure says the job disappears. The US Bureau of Labor Statistics projects employment of secretaries and administrative assistants to shrink 6 percent, from 1,889,800 jobs in 2025 to 1,775,700 in 2035. What AI changes first is tasks, not whole jobs: of the 8 core tasks listed for this work, 2 already show up in AI use data. The practical move either way: learn to direct AI on those tasks and check its work.
What to learn firstThe numbers, with sourcesStart free: lesson 1
The public measures, side by side
Each one answers a different question. None of them is a prediction that your job disappears, and we do not blend them into one score.
Jobs outlook to 2035
The US Bureau of Labor Statistics projects employment of secretaries and administrative assistants to shrink 6 percent, from 1,889,800 jobs in 2025 to 1,775,700 in 2035, with about 163,900 openings a year as people retire or change jobs.
What BLS itself says is changing this job:
- “Productivity change, share decreases as technology, including artificial intelligence (AI) systems, assists workers with routine back office functions, such as preparing documents, scheduling, and notetaking.”
What it does not mean: it is a ten year projection of total jobs, revised every year. It says nothing about any one employer or person.
How people use AI on this work
In Anthropic's Economic Index, conversations about this job's tasks were about 29 percent people working with the AI (learning, iterating, checking) and 71 percent handing a task over to it.
What it does not mean: it describes how people who already use one AI assistant use it, not how many jobs change.
How much of the work AI can assist
Microsoft Research rates this occupation 0.26 on its AI applicability scale of 0 to 1, from what people actually asked an AI assistant to help with.
What it does not mean: its own authors say applicability is not job loss. A high rating means AI can help with much of the work, which is a reason to learn it.
Where AI already shows up in this work
Core tasks of the job, from O*NET, that appear in real AI use. These are the parts of the day that change first.
- Create, maintain, and enter information into databases.
- Use computers for various applications, such as database management or word processing.
What stays with people
Core tasks with no AI use observed in the published data. That measures today, not a promise about tomorrow; these are where judgment, trust and presence carry the work now.
- Perform payroll functions, such as maintaining timekeeping information and processing and submitting payroll.
- Operate office equipment, such as fax machines, copiers, or phone systems and arrange for repairs when equipment malfunctions.
How long do I have?
Nobody can honestly name a year. The measured pattern so far is that hiring for junior roles slows first in the most exposed work, while people already in the job take on AI assisted tasks. The BLS projection above runs to 2035 and is revised every year.
Stanford's Digital Economy Lab found no widespread displacement, but employment for young workers in the most AI exposed jobs has fallen 19 percent behind comparable peers (August 2026). Read the study.
The practical answer is the same either way: the person who can direct AI on these tasks, and catch it when it is wrong, is the one a team keeps and hires.
What to learn first
In this order, for this job. Each is taught in GAGE's AI literacy program, Certified AI Practitioner: Workplace Foundations.
1.Use AI to keep databases and records current, and check what it enters
Taught in Module 4, Practical AI Workflow Design and Prompt Engineering; Module 7, Advanced AI Literacy
2.Build a small set of reusable prompts for the letters and notices you write every week
Taught in Module 4, Practical AI Workflow Design and Prompt Engineering
3.Set up filing so both people and AI tools can find things
Taught in Module 1, Digital Foundations
4.Keep payroll and money tasks under a check step you never skip
Taught in Module 3, Ethical and Responsible AI and Operational Governance; Module 5, Critical Thinking and Context Engineering
5.Spot fake calls and messages made with AI before they reach your boss
Taught in Module 10, AI Security Fundamentals
Show an employer you can do it
A course certificate says you finished. At GAGE you explain each idea back in your own words and it is graded against the lesson, so your record shows what you understood. An employer can check it with one link.
Related jobs
Questions
- Will AI replace secretaries and administrative assistants?
- No published measure says the job disappears. The US Bureau of Labor Statistics projects employment of secretaries and administrative assistants to shrink 6 percent, from 1,889,800 jobs in 2025 to 1,775,700 in 2035. What AI changes first is tasks, not whole jobs: of the 8 core tasks listed for this work, 2 already show up in AI use data. The practical move either way: learn to direct AI on those tasks and check its work.
- How long do secretaries and administrative assistants have before AI changes the job?
- Nobody can honestly name a year. The measured pattern so far is that hiring for junior roles slows first in the most exposed work, while people already in the job take on AI assisted tasks. The BLS projection above runs to 2035 and is revised every year.
- What should secretaries and administrative assistants learn about AI first?
- Use AI to keep databases and records current, and check what it enters; Build a small set of reusable prompts for the letters and notices you write every week; Set up filing so both people and AI tools can find things. GAGE teaches each of these in its AI literacy program, and the first lesson is free.
Sources
- U.S. Bureau of Labor Statistics, Employment Projections program, Employment Projections, Table 1.2: Occupational projections, 2025 to 35, and worker characteristics, 2025, 2025 to 2035 projections (tables page last modified August 27, 2026). The requested 2024 to 2034 set is superseded; the current table is used., Public domain (bls.gov/opub/copyright-information.htm: everything BLS publishes is in the public domain; cite BLS as the source). Read 2026-09-25.
- Anthropic, Anthropic Economic Index, sixth release (claude_ai, monthly aggregates), release_2026_06_26 (report: Anthropic Economic Index report: Cadences, 2026-06-26; dataset commit 2ea58ff75e4247d26810c37f10c179edc2466cac). Period read: GLOBAL, 2026-05-01 to 2026-06-01, the latest month in the file., CC BY (release_2026_06_26/data_documentation.md: Data released under CC-BY; the repository README says data CC-BY, code MIT. The Hugging Face card metadata says mit, which covers the code.). Read 2026-09-25.
- Microsoft Research (Tomlinson, Jaffe, Wang, Counts, Suri), Working with AI: Measuring the Applicability of Generative AI to Occupations (AI applicability scores), v1.1 (arXiv 2507.07935 revision v6), repository commit c94a07c52fb1d88ca5d221388f06d10e1bd6d2fe of 2025-12-22. Bing Copilot conversations, United States, 2024-01-01 to 2024-09-30., CC BY 4.0 (LICENSE.txt and README: This repository is licensed under CC BY 4.0). Read 2026-09-25.
- National Center for O*NET Development, for the U.S. Department of Labor, Employment and Training Administration, O*NET 31.0 Database: Task Statements and Task Ratings, O*NET 31.0 (August 2026, the current production release per onetcenter.org/db_releases.html on 2026-09-25), CC BY 4.0 (onetcenter.org/database.html: licensed under a Creative Commons Attribution 4.0 International License). Read 2026-09-25.
- This page includes information from O*NET OnLine 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. GAGE has modified all or some of this information; USDOL/ETA has not approved, endorsed, or tested these modifications.
- Occupation code 43-6014. The skills, task grouping and plain language on this page are GAGE's own.