Jobs / Science and research
Can agents do science and research work?
Life, physical, and social science research and technical work.
- US jobs
- 1,473,280 (0.9% of employment)
- Typical wage
- $88K a year (employment-weighted mean of occupation medians)
- Occupations
- 66 in O*NET 31.0
- Evidence
- 2 benchmarks from 2 institutions
Task level
Partial · thin evidence (fewer than 2 independent institutions)
Thin evidence. CORE-Bench asks agents to reproduce the results of published papers in computer science, social science, and medicine from the authors' code and data. Princeton's HAL team declared it solved in December 2025: Claude Code with Claude Opus 4.5 scored 95.5% after manual regrading. No benchmark in our collection tests new experiments or lab work.
Evidence: CORE-Bench — by Princeton University. Reading dated 23 Sep 2026.
Agents succeed on some scoped tasks. Reliable or end-to-end work is not shown.
Project level
Not yet · thin evidence (fewer than 2 independent institutions)
Thin evidence. On PaperBench, where agents replicate machine learning papers from scratch, the best official result is 26.0% (o1, April 2025). On a small subset, ML PhDs scored 41.4% against o1's 26.6%. No newer official results exist, and no benchmark tests research projects in other sciences.
Evidence: PaperBench — by OpenAI. Reading dated 23 Sep 2026.
The best agents fail most tasks on the best available benchmarks.
Benchmarks for this work
- 95.5%Claude Code + Claude Opus 4.5 (after manual regrading) · 3 Dec 2025Solved
Also relevant
These benchmarks cover part of this family's work.
Functions: Research replication
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.
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 science and research work?." https://canagentswork.com/jobs/science-research/ (accessed 2026-09-24). CC BY 4.0.@misc{caw-jobs-science-research,
title = {{Can agents do science and research work?}},
author = {{Can Agents Work}},
year = {2026},
howpublished = {\url{https://canagentswork.com/jobs/science-research/}},
note = {Accessed 2026-09-24. CC BY 4.0}
}