Physician assistants: AI exposure and career outlook
Physician assistants (SOC 29-1071) sit at the 6th percentile for measured AI exposure among the 342 occupations tracked here, measured from a composite of Microsoft Research and Anthropic Economic Index telemetry. An estimated 5% of tasks are already automated and 13% are being reshaped rather than replaced — both modelled figures, not direct measurements. Of the Claude usage observed in this occupation, 51% follows an automation pattern and 49% an augmentation pattern, which places it in the Insulated group: little measured AI exposure so far. No urgency here, though the tools are still worth having in your hands. In Canada the role maps to NOC 31303 (Physician assistants, midwives and allied health professionals), and ESDC's COPS 2024–2033 outlook for that unit group is strong risk of shortage.
Figures last updated 2026-06. Every number on this page is labelled measured or modelled; where a source has no coverage for this occupation, it says so rather than showing a zero.
Key facts
- SOC code
- 29-1071
- Canada NOC 2021
- 31303
- TEER level
- 1 — University Degree
- COPS outlook
- Strong risk of Shortage
What the measurements actually say
Three independent studies measure AI exposure for this occupation directly. The two percentages below them are ours, modelled from those measurements.
| Metric | Value | Provenance |
|---|---|---|
| AI applicability | 5% | Measured — Microsoft Research, from 200,000 Copilot conversations classified against O*NET work activities. |
| Observed AI usage | 0% | Measured — Anthropic Economic Index, share of tasks observed being performed with Claude (CC-BY 4.0). |
| Academic AI exposure | 54th percentile | Measured — Felten-Raj-Seamans AIOE index across 774 occupations; the index underlying Statistics Canada's Canadian estimates. |
| Automation-pattern usage | 51% | Measured — the share of observed Claude usage in this occupation where the task is handed over rather than iterated on (Anthropic Economic Index). |
| Augmentation-pattern usage | 49% | Measured — the complement of the row above; the two sum to 100% by construction. This is the pattern where the person stays in the loop. |
| Estimated task automation (modelled) | 5% | Modelled from this occupation's measured telemetry composite, mapped through anchors calibrated to Anthropic's published aggregate findings. Our estimate, not per-occupation telemetry. |
| Estimated task reshaping (modelled) | 13% | Share of tasks where AI acts as co-pilot rather than replacement. Modelled from the measured telemetry composite, calibrated to BCG's published aggregate reshaping rates. Our estimate. |
This skill domain has minimal AI substitution risk. AI is primarily augmenting — not replacing — professionals here.
In Canada: NOC 31303
Physician assistants map to NOC 31303 — Physician assistants, midwives and allied health professionals at TEER 1 (University Degree). ESDC's COPS 2024–2033 projection for this unit group is Strong risk of Shortage.
Mapped via Statistics Canada's official SOC 2018 → NOC 2016 → NOC 2021 correspondence tables (closest match of several correspondence candidates). Several candidates were available, and this one matched the occupation title on every word — no other did — so the TEER level read off its second digit is reported.
What this job actually involves
These are the O*NET task statements for Physician assistants— the work itself, as reported by people doing the job — grouped by the work activity each task belongs to. Showing 12 of 28, core tasks first.
The exposure figures above are for the occupation as a whole. We do not have a per-task measurement: no published dataset tells us which of these specific tasks AI is doing. Read this as what the job is made of, not as a ranking of what is at risk.
2 of these 12statements do carry one measured signal: Anthropic publishes, per O*NET task, how much of the observed Claude usage on it looked like automation rather than iteration. Those are marked below. It still says how people use AI on that task, not how much of it AI can do — and the unmarked statements carry nothing rather than inheriting the occupation's figure.
Microsoft's applicability score was itself produced by classifying Copilot conversations against the O*NET work-activity catalogue — the same catalogue these groups come from. The published file gives one number per occupation, not one per activity, so we can tell you that 5% of this occupation's work activities are covered by observed AI usage, but not which ones.
Assisting and Caring for Others · 3 tasks
Provide airway management interventions including tracheal intubation, fiber optics, or ventilary support.
Core task · Importance 5.0/5 (O*NET)
Administer anesthetic, adjuvant, or accessory drugs under the direction of an anesthesiologist.
Core task · Importance 5.0/5 (O*NET)
Respond to emergency situations by providing cardiopulmonary resuscitation (CPR), basic cardiac life support (BLS), advanced cardiac life support (ACLS), or pediatric advanced life support (PALS).
Core task · Importance 4.8/5 (O*NET)
Inspecting Equipment, Structures, or Materials · 2 tasks
Examine patients to obtain information about their physical condition.
Core task · Importance 4.9/5 (O*NET)
Pretest and calibrate anesthesia delivery systems and monitors.
Core task · Importance 4.8/5 (O*NET)
Monitoring and Controlling Resources · 2 tasks
Prescribe therapy or medication with physician approval.
Core task · Importance 4.9/5 (O*NET)
Administer or order diagnostic tests, such as x-ray, electrocardiogram, and laboratory tests.
Core task · Importance 4.9/5 (O*NET)
Getting Information · 1 task
Obtain, compile, and record patient medical data, including health history, progress notes, and results of physical examination.
Core task · Importance 4.9/5 (O*NET)
Monitoring Processes, Materials, or Surroundings · 1 task
Assist anesthesiologists in monitoring of patients, including electrocardiogram (EKG), direct arterial pressure, central venous pressure, arterial blood gas, hematocrit, or routine measurement of temperature, respiration, blood pressure or heart rate.
Core task · Importance 5.0/5 (O*NET)
Analyzing Data or Information · 1 task
Interpret diagnostic test results for deviations from normal.
Core task · Importance 4.9/5 (O*NET) · 48% of observed AI use on this task is automation-pattern
Making Decisions and Solving Problems · 1 task
Make tentative diagnoses and decisions about management and treatment of patients.
Core task · Importance 4.9/5 (O*NET) · 71% of observed AI use on this task is automation-pattern
Handling and Moving Objects · 1 task
Control anesthesia levels during procedures.
Core task · Importance 5.0/5 (O*NET)
Task statements and work activities from the O*NET Database by the U.S. Department of Labor, Employment and Training Administration (USDOL/ETA), used under CC BY 4.0. O*NET® is a trademark of USDOL/ETA. USDOL/ETA has not reviewed or approved this material.
That's the average. Is this you?
Everything above describes Physician assistants as a whole. Three questions, and nothing is stored or sent.
Where are you in your career?
What does your week mostly look like?
Is your employer deploying AI?
Where the leverage is
With low AI exposure, the focus for this role is on adopting productivity tools. Leverage AI for scheduling, administrative tasks, and analytics to double operational throughput.
Physical requirements and manual dexterity provide strong natural insulation from AI substitution. AI is primarily a peripheral tool for diagnostics.
Upskill in AI prompt engineering, model validation, and governance. Focus on transition playbooks to pivot toward high-value advisory services.
This guidance is written per exposure band — three texts across all 342 occupations, framed on the ILO's transformation-versus-displacement distinction. It is editorial guidance for the safe band, not an occupation-specific research finding about physician assistants.
Sources for this occupation
Only the sources that hold data for physician assistants are listed. A study that does not cover this occupation is not cited here.
- Microsoft Research — Working with AI (Tomlinson et al., arXiv:2507.07935): Measured AI applicability, from 200,000 anonymised Copilot conversations classified against O*NET work activities.
- Anthropic Economic Index: Measured observed AI usage per occupation, from Anthropic's open per-SOC dataset (CC-BY 4.0).
- Felten, Raj & Seamans — AI Occupational Exposure index: Measured academic exposure percentile; the index underlying Statistics Canada's Canadian AI-exposure estimates (Mehdi & Morissette, 2024).
- O*NET Database 31.0 (USDOL/ETA), CC BY 4.0: The occupation's task statements and work activities, verbatim. Joined on SOC 2018 with no crosswalk — O*NET-SOC is built on SOC — and carrying no claim about which tasks AI touches.
- Anthropic Economic Index — collaboration split (release 2026-06-26, CC-BY 4.0): Measured share of observed Claude usage following an automation rather than an augmentation pattern. This is the second axis of the grouping above.
- Statistics Canada NOC 2021 concordance: Official SOC 2018 → NOC 2016 V1.3 → NOC 2021 V1.0 correspondence tables. TEER is the second digit of the resolved code, so we report it only where the concordance gave a single candidate.
- ESDC COPS 2024–2033: Projected labour-market assessment per NOC 2021 unit group, from the Canadian Occupational Projection System open dataset.
- BCG — AI Will Reshape More Jobs Than It Replaces (April 3, 2026): Published aggregate reshaping rates, used to calibrate our modelled percentages. The per-occupation figures here are our estimates, not BCG's data.
Related occupations
Other roles in Trades & Healthcare.
This is the picture for the role. What about your skills?
These numbers describe an occupation, not a person. Map your own skills against them, see which pathways stay open, and plan the transition you actually want.