Physician Assistant (PA)
Physician assistant work fundamentally resists automation because clinical medicine requires integrating information that cannot be fully captured in data.
In real usage data, AI shows up in this job less often than in most occupations.
Technical detail
Based on Microsoft's "Working with AI" study of real Copilot conversations mapped to O*NET tasks (arXiv 2507.07935): Physician Assistants scored 0.051 on AI applicability, notably below the cross-occupation mean (0.159, stdev 0.098) across all 785 SOC codes studied — banded here as Low real-world AI usage relative to other occupations.
What a physician assistant (pa) actually does
Physician assistants practice medicine under the supervision of physicians, examining patients, diagnosing illnesses, developing treatment plans, prescribing medications, and performing procedures. The work encompasses taking patient histories, conducting physical examinations, ordering and interpreting diagnostic tests, counseling patients on preventive care and treatment options, assisting in surgery, and managing chronic disease care. PAs work across medical specialties including primary care, emergency medicine, surgery, psychiatry, dermatology, and orthopedics, often serving as the primary care provider for many patients.
The role requires clinical decision-making skills to diagnose conditions based on symptoms and test results, communication skills to explain complex medical information to patients, and the judgment to recognize when cases require physician consultation. PAs in primary care settings often manage patient panels independently, providing continuity of care that builds long-term therapeutic relationships. In surgical settings, they assist with procedures and manage pre- and post-operative care. The profession requires graduate-level education and national certification.
Why it reads this way
Physician assistant work fundamentally resists automation because clinical medicine requires integrating information that cannot be fully captured in data. A PA examining a patient synthesizes verbal complaints, physical findings, medical history, social context, and subtle observations—the patient's affect, their family dynamics, whether they seem to be minimizing symptoms. Diagnosis involves pattern recognition refined through thousands of patient encounters, intuition about when something doesn't fit typical presentations, and judgment about which possibilities to investigate.
Treatment decisions require balancing clinical evidence with patient preferences, considering factors like medication costs, lifestyle constraints, and what patients will actually do. The therapeutic relationship itself has healing power; patients who trust their PA are more likely to follow treatment plans and disclose important information. Physical examination requires hands-on assessment—palpating for tenderness, listening to heart and lung sounds, observing gait and movement—that AI cannot perform.
The judgment to recognize when a presentation requires urgent intervention versus watchful waiting remains distinctly human.
Skills this role draws on
We don't yet have task-by-task time-share data for this occupation, so we can't show which specific tasks carry the exposure score above. This is the real skill set instead.
Typical salary range (USD)
$95,000–$182,000
Average: $141,280/yr
Range source: BLS Occupational Employment Statistics, May 2024
Average source: BLS Occupational Employment and Wage Statistics, May 2025 (Annual Mean Wage)
Related careers
Head-to-head comparisons
This occupation's structuralScore of 65 clears the Safe threshold, and a lightweight robotics/autonomous-systems check found no material physical-automation threat, so the standard rubric applies without an override. As of August 2026, all 50 states require PA licensure tied to the individual practitioner, and a February 2026 Nature Medicine study found a leading consumer health AI tool under-triaged 52% of medical emergencies, directly undercutting the case that generative AI can safely replace a PA's diagnostic judgment today. A narrow, FDA-cleared autonomous diagnostic tool (LumineticsCore, cleared since 2018 for diabetic retinopathy screening) shows bounded autonomous clinical AI is technically possible, but it has not expanded into general PA-scope tasks in the eight years since clearance. This occupation remains Safe.
This is a researched judgment (checklist v1.2), not a statistical measurement — it sits alongside, not instead of, the structural exposure signal above. Next scheduled review: 2027-08-10.
Sources (5)
- https://www.bls.gov/ooh/healthcare/physician-assistants.htm
- https://www.ama-assn.org/sites/ama-assn.org/files/corp/media-browser/public/arc-public/state-law-physician-assistant-scope-practice.pdf
- https://www.fda.gov/regulatory-information/search-fda-guidance-documents/clinical-decision-support-software
- https://www.digitaldiagnostics.com/fda-permits-marketing-of-lumineticscore-formerly-known-as-idx-dr-for-automated-detection-of-diabetic-retinopathy-in-primary-care/
- https://www.nature.com/articles/s41591-026-04297-7
Data sources & methodology
Salary data: BLS Occupational Employment Statistics, May 2024. Min/max figures represent the 10th–90th percentile annual wage range across the United States. Average figure sourced separately: BLS Occupational Employment and Wage Statistics, May 2025 (Annual Mean Wage).
Automation Risk Score: Based on O*NET occupational analysis (29-1071.00) evaluating task complexity, physical requirements, social intelligence, and environmental variability. Methodology based on research from Frey & Osborne (Oxford, 2017).
Growth projections: 20% (2024-2034), based on BLS Occupational Outlook Handbook.