100×worker · analysis
Which jobs will AI replace?
In 2025 Microsoft Research published the most grounded answer available to this question. Instead of asking experts to speculate, they took 200,000 real Copilot conversations and mapped what people actually asked AI to do against the official work-activity lists of 786 occupations. The result is an "AI applicability score" per occupation: the share of its work activities where generative AI is demonstrably usable today. Anthropic's Economic Index, built on millions of Claude conversations, points the same way.
Both datasets agree on the core finding: AI lands on tasks, not on jobs. A job survives as long as part of its bundle stays human (accountability, trust, physical work, hard conversations). What changes is the ratio of executing to directing and reviewing.
The 30 most exposed occupations (Microsoft Research, 2025)
| # | Occupation | AI applicability | |
|---|---|---|---|
| 1 | Interpreters and Translators | 49% | Task breakdown |
| 2 | Historians | 46% | |
| 3 | Writers and Authors | 45% | Task breakdown |
| 4 | Sales Representatives of Services, Except Advertising, Insurance, Financial Services, and Travel | 45% | |
| 5 | Computer Numerically Controlled Tool Programmers | 42% | |
| 6 | Broadcast Announcers and Radio Disc Jockeys | 41% | |
| 7 | Customer Service Representatives | 41% | Task breakdown |
| 8 | Telemarketers | 40% | |
| 9 | Political Scientists | 39% | |
| 10 | Mathematicians | 39% | |
| 11 | News Analysts, Reporters, and Journalists | 38% | |
| 12 | Passenger Attendants | 38% | |
| 13 | Technical Writers | 37% | |
| 14 | Concierges | 37% | |
| 15 | Proofreaders and Copy Markers | 37% | |
| 16 | Editors | 37% | |
| 17 | Business Teachers, Postsecondary | 37% | |
| 18 | Public Relations Specialists | 37% | |
| 19 | Data Scientists | 36% | |
| 20 | Personal Financial Advisors | 36% | Task breakdown |
| 21 | Web Developers | 35% | |
| 22 | Advertising Sales Agents | 35% | |
| 23 | Management Analysts | 35% | Task breakdown |
| 24 | Geographers | 35% | |
| 25 | Brokerage Clerks | 35% | |
| 26 | Market Research Analysts and Marketing Specialists | 35% | Task breakdown |
| 27 | Economics Teachers, Postsecondary | 35% | |
| 28 | Public Safety Telecommunicators | 35% | |
| 29 | Counter and Rental Clerks | 34% | |
| 30 | Telephone Operators | 34% |
The 15 least exposed occupations
| # | Occupation | AI applicability |
|---|---|---|
| 1 | Dredge Operators | 0% |
| 2 | Bridge and Lock Tenders | 0% |
| 3 | Water and Wastewater Treatment Plant and System Operators | 0% |
| 4 | Foundry Mold and Coremakers | 0% |
| 5 | Rail-Track Laying and Maintenance Equipment Operators | 0% |
| 6 | Pile Driver Operators | 0% |
| 7 | Floor Sanders and Finishers | 0% |
| 8 | Orderlies | 0% |
| 9 | Motorboat Operators | 0% |
| 10 | Logging Equipment Operators | 1% |
| 11 | Paving, Surfacing, and Tamping Equipment Operators | 1% |
| 12 | Ophthalmic Medical Technicians | 1% |
| 13 | Maids and Housekeeping Cleaners | 1% |
| 14 | Roustabouts, Oil and Gas | 1% |
| 15 | First-Line Supervisors of Firefighting and Prevention Workers | 1% |
Source: Microsoft Research, "Working with AI: Measuring the Applicability of Generative AI to Occupations" (2025), data public under CC BY 4.0 at github.com/microsoft/working-with-ai. The score measures task exposure, not job-loss probability.
Jobs we have analyzed task by task
For these occupations we went one level deeper: every task sorted into eliminate, automate, delegate, or keep, with the moves that make you the person who runs AI on your team.
Why "replace" is the wrong frame
An interpreter scores 49% and a registered nurse 12% for one reason: translation is almost entirely language work, which is exactly what large language models do, while nursing is mostly physical care and human contact. Both jobs will exist in ten years. Only the translator's work week will look fundamentally different, with AI producing first drafts and the human doing quality, nuance, and accountability.
That is the pattern across all 786 occupations: the score predicts how much of your week changes, never whether you are needed. The people at risk are those who keep executing tasks AI already does well. The people who gain are those who learn to direct, review, and improve AI work first. Our per-job analyses show what that means concretely, and the free task scan maps it for your specific role.
Frequently asked questions
Which jobs will AI replace?
Very few in their entirety. AI takes over tasks, not occupations. Language- and information-heavy roles see the largest share of their task list shift; physical and relational work the least.
What does the AI applicability score measure?
The share of an occupation's work activities where generative AI is demonstrably usable today, measured from real Copilot usage. Task exposure, not job-loss probability.
What should I do if my job scores high?
Map your own week, sort each task into eliminate, automate, delegate, or keep, and build an AI working method for the middle two. Being first at this makes you more valuable, not less.
Where does your job stand?
The free scan builds your personal task map in 30 seconds, based on your role and industry.
Run the free task scan