AI Delegation Index

AI Delegation Index

What work could you hand off to an AI agent?

Physicists

Strong

AI agents could handle several recurring digital workflows in this job, while people remain responsible for judgment and final decisions.

Where agents can help most

  1. 1

    Validate results for papers and talks

    Use an agent to organize experimental results, convert notes and tables into a draft results section, and check equations, units, and reported values against the dataset.

  2. 2

    Compare simulations with theory

    Use an agent to gather simulation inputs, run comparison summaries against measured data, and lay out where the model matches or misses across different conditions.

  3. 3

    Review papers for method issues

    Use an agent to read a submitted paper, pull out the key equations, data checks, and method details, and draft a review memo with possible strengths and concerns.

Search another job

O*NET-SOC 19-2012.00 · #75 of 923

Result context

How to read this result

Conduct research into physical phenomena, develop theories on the basis of observation and experiments, and devise methods to apply physical laws and theories.

National position
#75 of 923 occupations
Top 9% of occupations
Overall AI delegation potential
Strong
Potential of score-contributing workflows
Strong · 82/100
Meaningful work covered
72%

The overall rating combines how useful the best agent workflows are with how much of the occupation they address. It is not an estimate of job automation or replacement.

Recommended agent uses

3 workflows you could delegate to AI

1

Validate results for papers and talks

How you could use an agent

Use an agent to organize experimental results, convert notes and tables into a draft results section, and check equations, units, and reported values against the dataset. It can help prepare a cleaner package for a paper or conference talk so you can focus on the scientific interpretation and final wording.

Where you stay involved

You decide what the data really supports, revise the technical claims, and approve anything that will be shared outside the group. You remain responsible for the conclusions, the math, and any safety-sensitive or contested interpretation.

Review level: Medium

2

Compare simulations with theory

How you could use an agent

Use an agent to gather simulation inputs, run comparison summaries against measured data, and lay out where the model matches or misses across different conditions. It can track parameter changes and write up the differences so you can decide what assumptions to adjust next.

Where you stay involved

You judge whether the model needs a new assumption, a parameter change, or a deeper rethink of the theory. You also decide when the mismatch is significant enough to bring to senior researchers.

Review level: Medium

3

Review papers for method issues

How you could use an agent

Use an agent to read a submitted paper, pull out the key equations, data checks, and method details, and draft a review memo with possible strengths and concerns. It can help you compare the manuscript’s claims with the underlying analysis and with familiar references so your review is faster to write.

Where you stay involved

You decide whether the methods are sound, whether the paper needs revision, or whether the journal should reject it. You also handle any concern about misconduct, safety, or conflicts of interest through the proper human channel.

Review level: Medium

Supporting analysis

Why this occupation's AI delegation potential is Strong

Underlying methodology score

77 / 100

This technical score determines the qualitative rating; it is not an estimate of the share of the occupation that can be automated.

Importance & frequency80
AI capability86
Digital actionability90
End-to-end leverage76
Safety & reversibility75
Meaningful-work coverage
72%
Physical-work modifier
Limited
Safety modifier
Limited
Qualitative judgment
No material constraint
O*NET task evidence
16 tasks

O*NET 31.0 · methodology 3.3.0. Every workflow passes an action-level physical-execution and protected human-and-veterinary clinical-action gate. Documentation workflows must own a complete digital loop and use digital task evidence only; support-only workflows are disclosed separately and excluded from scoring. Artistic, editorial, normative, and policy-dependent work receives a transparent human-judgment constraint. National ranking within 923 scored O*NET occupations under methodology 3.3.0.

Useful comparisons

Similar occupations

Compare side by side

Economists

AI delegation potential: Moderate

Top 23% of occupations · #211

Historians

AI delegation potential: Strong

Top 10% of occupations · #87

Survey Researchers

AI delegation potential: Strong

Top 7% of occupations · #56

Atmospheric and Space Scientists

AI delegation potential: Strong

Top 9% of occupations · #80