Industry group, 5417
Scientific Research and Development Services
931,790 people work in this industry in the US, at mean pay of $136,980 a year (BLS, May 2025). What follows is how much of their working time the best AI models could finish, when, and in which jobs and tasks. Estimates.
of the paid hours. Now is projected from METR's April 2026 measurement and Epoch's November 2026 index.
a year: 723 million hours at what they cost. Not a forecast of savings.
- 67%
- 17%
- 16%
of the paid hours are within reach now in tasks people already bring to Claude. Another 27% is within reach in tasks not yet seen there.
Across the US: 30% within reach now and 36% by the end of 2028, on the long-run pace at four in five. Now is projected from METR's April 2026 measurement and Epoch's November 2026 index.
Share of the paid hours within reach, now to 2030
Change assumptionsHide assumptionsLong-run pace, four in five
What each choice means: , , , , . Every pace starts from .
What handing the work over would take
For the 750,148 people in jobs so far, 81% of the industry's people; the others are in the totals above and join this split as the review reaches their jobs. Shares of their working time within reach now, at the pace and reliability chosen above. Estimates. Now is projected from METR's April 2026 measurement and Epoch's November 2026 index.
about 501 million hours a year. The material can be handed to a model and the result handed back.
about 119 million hours a year. The model has to be connected to the system the work lives in.
about 181 million hours a year. Short enough by length. This site leaves conversation with people, so it is not in the totals.
What their work is made of
- 57%
- 9%
- 15%
- 5%
- 14%
Each task's time split among five kinds of doing, judged one task at a time; only documents and systems can count as within reach.
What changes for the people
For the 750,148 people in jobs reviewed task by task so far, by the end of 2028 on the long-run pace at four in five: the shares of their working time, from each job page's own map, weighted by how many people here do that job. It says which parts of the work change, never whether anyone keeps a job. Estimates.
Conversation, in-person and hands-on work.
Desk work within reach in tasks where a mistake costs something or a licensed person signs.
Desk work within reach that is small if wrong, with lighter checks.
What most people here would learn first
How many of them have each skill among the first two things to learn on their job's page. Every state, and the US
- Reviewing work you did not do yourself652,891 (87%)
- Briefing a model and checking its drafts573,596 (76%)
- The hands-on craft95,445 (13%)
- Judgment where errors are severe69,900 (9%)
- Running AI inside the business systems40,970 (5%)
- The conversations38,737 (5%)
- Dealing with people outside the organization13,000 (2%)
- Signing off under your license10,227 (1%)
- Handling protected records350 (under 1%)
How far this has come
With GPT-4, measured in March 2023, 0% of this industry's paid hours were within reach at . With Claude 3.7 Sonnet, measured in February 2025, 0%. With Claude Mythos Preview (early), measured in April 2026, 12%. Today's estimate: 38%.
A backcast: today's task data with the best model METR had measured by each date. It shows how fast the models have moved, not what anyone forecast at the time.
Robots and the physical work
Anthropic rates 20% of the working time of the people in this industry as physical work. For 16% of the working time, a robot has been shown doing most of the task somewhere: 13% in a purpose-built cell, 3% in a structured facility. 0.19 of 3.
- Not physical 80%
- No robot shown 4%
- In a purpose-built cell 13%
- In a structured facility 3%
Anthropic's rating, not this site's estimate: Claude Opus 5, searching the web, rated each physical O*NET task by the kind of setting in which a robot it could cite (deployed, sold or demonstrated) does most of the task, weighted by its own estimate of each task's share of the time (What work can robots do?, September 30, 2026, data CC BY 4.0; every job ranked). A robot shown somewhere is not a robot in your building: cost, setup and safety decide that. The share within reach on this page still counts no body work. Weighted by the people in each occupation; each job's page names the robot behind each rated task.
Where the hours are
Jobs by hours within reach now. Now is projected from METR's April 2026 measurement and Epoch's November 2026 index. The 40 largest jobs hold 62% of the people.
| People | Pay here | |||
|---|---|---|---|---|
| Software Developers | 50,270 | 48% | 50 million | $158K |
| General and Operations Managers | 31,740 | 45% | 30 million | $201K |
| Project Management Specialists | 24,420 | 58% | 29 million | $127K |
| Natural Sciences Managers | 44,430 | 31% | 28 million | $209K |
| Industrial Engineers | 16,060 | 60% | 20 million | $127K |
| Biological Technicians | 25,090 | 35% | 18 million | $62K |
| Data Scientists | 11,480 | 71% | 17 million | $131K |
| Computer and Information Systems Managers | 16,160 | 45% | 15 million | $216K |
| Secretaries and Administrative Assistants, Except Legal, Medical, and Executive | 11,370 | 59% | 14 million | $61K |
| Social Science Research Assistants | 9,260 | 72% | 14 million | $63K |
| Office Clerks, General | 11,010 | 56% | 13 million | $49K |
| Compliance Officers | 11,470 | 54% | 13 million | $107K |
People: BLS counts for this industry, May 2025. Pay here: the median paid for the job in this industry, else its mean, else the national median. 329 smaller jobs (345,240 people) count in every total but are not listed. Task data covers 99% of the industry's people; the rest are in BLS "all other" groups with no task list, or in counts BLS does not publish.
The tasks with the most hours within reach by the end of 2028
Across the whole industry, on the chosen pace and reliability.
- Design, develop and modify software systems, using scientific analysis and mathematical models to predict and measure outcomes and consequences of design.Software Developers17 million hours80% of the task within reach
- Write and publish articles in scientific journals.Medical Scientists, Except Epidemiologists15 million hours98% of the task within reach
- Modify existing software to correct errors, adapt it to new hardware, or upgrade interfaces and improve performance.Software Developers14 million hours100% of the task within reach
- Write applications for research grants.Medical Scientists, Except Epidemiologists14 million hours96% of the task within reach
- Analyze user needs and software requirements to determine feasibility of design within time and cost constraints.Software Developers9.8 million hours70% of the task within reach
- Conduct own research in field of expertise.Natural Sciences Managers9.5 million hours65% of the task within reach
- Develop or direct software system testing or validation procedures, programming, or documentation.Software Developers9.1 million hours80% of the task within reach
- Design or coordinate successive phases of problem analysis, solution proposals, or testing.Natural Sciences Managers8.9 million hours70% of the task within reach
- Develop computer information resources, providing for data security and control, strategic computing, and disaster recovery.Computer and Information Systems Managers8.4 million hours92% of the task within reach
- Study animal and human health and physiological processes.Medical Scientists, Except Epidemiologists8.1 million hours55% of the task within reach
Scan a typical company in this industry
The industry's largest jobs, in the proportions BLS counts across the whole industry, at the pay it pays, run through the company scan. A starting point before you have a roster, or a benchmark for the one you have. Or choose the whole industry: every occupation BLS publishes for it with task data, at its nationwide headcount, the sector report ($249 for the full report, or included in Pro).
Next to this industry
Within reach now and by the end of 2028, long-run pace, four in five. Now is projected from METR's April 2026 measurement and Epoch's November 2026 index.
- Computer Systems Design and Related Services2,445,140 people50% now, 74% by 2028
- Management, Scientific, and Technical Consulting Services1,872,730 people47% now, 68% by 2028
- Architectural, Engineering, and Related Services1,719,390 people43% now, 68% by 2028
- Legal Services1,224,710 people52% now, 68% by 2028
- Accounting, Tax Preparation, Bookkeeping, and Payroll Services1,091,370 people56% now, 77% by 2028
- Other Professional, Scientific, and Technical Services881,820 people35% now, 44% by 2028
- Advertising, Public Relations, and Related Services481,600 people48% now, 64% by 2028
- Specialized Design Services151,930 people52% now, 68% by 2028
How it is built: BLS counts the people in each job in this industry; each job's tasks and their lengths give its share within reach at a date; the industry's share is those shares weighted by people. Value is pay times 1.43 for benefits and payroll costs. Seen in Claude use splits today's share by whether Anthropic saw people bring each task to Claude in 2025; other AI tools are not counted. How it works
Questions about this industry
- How much of the work in Scientific Research and Development Services is within reach of AI now?
- An estimated 38% of the paid hours in Scientific Research and Development Services are within reach of AI models now and 64% by the end of 2028, on METR's long-run pace at four in five (56% to 66% across the three paces the site shows). Within reach means a model could finish work that long at that reliability. It is not a forecast of jobs lost or of savings.
- Which jobs in Scientific Research and Development Services carry the most hours within reach?
- By hours within reach now (projected from METR's April 2026 measurement and Epoch's November 2026 index): Software Developers (50,270 people, 48% of their time, about 50 million hours a year); General and Operations Managers (31,740 people, 45% of their time, about 30 million hours a year); and Project Management Specialists (24,420 people, 58% of their time, about 29 million hours a year).
- Which tasks in Scientific Research and Development Services carry the most hours within reach by the end of 2028?
- Design, develop and modify software systems, using scientific analysis and mathematical models to predict and measure outcomes and consequences of design (Software Developers, about 17 million hours a year); Write and publish articles in scientific journals (Medical Scientists, Except Epidemiologists, about 15 million hours a year); and Modify existing software to correct errors, adapt it to new hardware, or upgrade interfaces and improve performance (Software Developers, about 14 million hours a year).
- What would people working in Scientific Research and Development Services learn first?
- Among the 750,148 people in jobs reviewed task by task, the skills most often among the first two things to learn on their job's page are reviewing work you did not do yourself (652,891 people); briefing a model and checking its drafts (573,596 people); and the hands-on craft (95,445 people). By the end of 2028 on the long-run pace at four in five, about 33% of their working time stays with people, 57% becomes checking a model's work and 7% can be handed over with lighter checks. It says which parts of the work change, never whether anyone keeps a job.
- How is this estimated?
- BLS counts the people in each occupation in Scientific Research and Development Services (May 2025); each occupation's share of working time within reach comes from its O*NET tasks, each with an estimated length checked against GDPval, and METR's measurements of how long a task the best AI models can finish, carried forward to the date from the latest, Claude Mythos Preview (early) in April 2026. The industry's share is those shares weighted by people. Task data covers 99% of its people, and 81% are in occupations reviewed task by task. Data checked November 10, 2026.
Embed these figures on your site
A small card with the share of working time within reach now and at the end of 2027, 2028 and 2030, the date and a link back. Free under CC BY 4.0; the card carries the credit.