SAFE
PAs spend their day doing physical exams, suturing, injections, casting, assisting in surgery, and delivering diagnoses face to face — none of which a language model or current robot performs. The automatable slice is real but bounded: chart notes, differential lists, order sets, patient instructions, and coding are already being drafted by AI, which compresses documentation time rather than the clinical visit. State licensure plus prescriptive authority means a liable human signs the chart and the prescription.
Headcount grew steadily across the period.
Median pay $112,260 → $135,880 -3.2% in real terms
This line is counted by the Bureau of Labor Statistics — the one figure on this page that isn't a judgement of ours. Headcount moves on demand, offshoring, demographics and the business cycle, and automation is one term among several, often not the loudest.
So a falling line is not evidence that AI did it, and a rising one is not evidence that it won't. Both happen in this register: some occupations resist automation and shrink anyway, others are highly automatable and keep growing. The marked year is 2020.
BLS projection, 2024–2034
+20.4% 162,700 → 195,800 on the projections basis
Hard to automate, and growing
The work resists current AI and the BLS projects +20.4% more of these jobs by 2034. Note that safe does not mean well paid — several of the fastest-growing resistant occupations are among the lowest paid on the register.
Different clocks. The score is what current AI could do to this work today. The projection is how many of these jobs will exist in 2034. Everything between the two — how fast employers actually adopt, whether demand grows in the meantime — is why they can point opposite ways without either being wrong.
~12,000 openings a year on average, including replacing people who leave.
Doctor AssistantAnesthesia AssistantAnesthetic AssistantAnesthesia SpecialistAnesthesia TechnicianPhysician's AssistantUrgent Care SpecialistAnesthesia TechnologistGynecological AssistantPhysician Assistant (PA)Physician Associate (PA)Operating Room TechnicianAnesthesiologist AssistantFamily Physician AssistantMedical Service TechnicianPediatric Physician AssistantCardiology Physician AssistantOrthopedic Physician AssistantCertified Anesthesia TechnicianOrthopaedic Physician AssistantAdvanced Practice Provider (AAP)Radiology Practitioner AssistantFamily Practice Physician AssistantCardiothoracic Anesthesia Technician
Holding it up: embodiment . Weakest point: task resistance .
Mixed — a routine tier and a judgment tier 13 rather than 16 because a real fraction of a PA's shift is genuinely displaceable — ambient scribes now draft the H&P and SOAP note, order sets and refill authorizations are protocol-driven, and much of urgent-care triage (strep, UTI, med refills) is already being routed to algorithmic e-visits — but the palpation, wound closure, joint injection, and pelvic exam that anchor the visit stay human.
Hands-on in uncontrolled environments 17 reflects that PA work happens on live bodies in uncontrolled settings — draining an abscess in a clinic room, first-assisting with retraction in an OR, reducing a dislocated shoulder in an ED bay, splinting on an ortho floor — with the remaining 3 points held back only because some PAs practice in narrow specialties (derm reads, psych med management, telehealth) where the hands-on share drops.
Licensed human required and personally liable 17 because PAs hold a state licence plus DEA registration and prescriptive authority, are named defendants in malpractice suits in their own right, and their signature on the chart and script is the legally operative act — short of 20 only because most states still tie practice to a collaborating or supervising physician agreement, so the accountability is shared rather than solely theirs.
Exists to be accountable for ambiguous calls 15 because the PA decides who gets a CT versus reassurance, who is admitted versus sent home, when chest pain is reflux and when it is an NSTEMI, and titrates opioids and antibiotics under stewardship pressure — high-stakes and ambiguous, though tempered by guidelines, standing orders, and the option to escalate to the supervising physician.
The verdict above describes this occupation as a whole. Almost nobody does the typical version of a job — tick what's actually in your week and see how your own mix sits.
Your task mix speaks to task resistance (13/20 here) — how much of the day's work current AI already does. That is the dimension the boxes above are about.
It cannot move the other three. Liability shield (17/20) is whether the law requires a licensed human to sign. Trust premium (15/20) is whether buyers specifically pay for a person. Judgment and accountability (15/20) is whether the role exists to own consequential calls. Those are facts about the occupation's standing, not about which tasks are in your week — a paralegal who does only trial exhibits still holds no licence. Together they are 47 of this occupation's 77 points (61%).
Embodiment (17/20) is also a property of the work rather than the worker, but we don't tag individual tasks as physical or not, so the picker can't tell you anything about it. That's a limit of this tool, not a claim.
Did we get the list right? Tell us what's missing — the tasks are written from the outside, and you're reading this from the inside.
The moves above are yours to make. This is the other half: what would have to change in the world for the occupation itself to score higher. None of it is in any one person's gift, but it is where the floor actually comes from. Scores here are not a one-way ratchet. Only two of the five dimensions — task resistance and embodiment — track what machines can do. The other three track law, what buyers will pay for, and who is answerable, and those move in both directions, often in response to the same pressure AI creates. If every lever below landed, this occupation would score around 86/100, still SAFE.
Continued expansion of PA full practice authority / optimal team practice (enacted in some form in Utah, North Dakota, Wyoming, Colorado 2024) so PAs own the disposition decision without physician co-signature, making them the accountable decision-maker of record for admit/discharge, escalation, and prescribing under ambiguity.
State medical/PA boards or legislatures explicitly requiring that AI-generated diagnostic impressions, order sets, and prescriptions be reviewed and signed by a licensed clinician who retains personal liability — the pattern in state 'AI in utilization review' laws already enacted (e.g., California SB 1120, 2024, barring AI-only denials) extended to point-of-care documentation and prescribing. Also DEA rules keeping controlled-substance prescribing tied to a licensed prescriber's DEA number.
Task-mix shift: as ambient scribes (Abridge, Nuance DAX) and AI coding absorb documentation and billing, the residual day is exam, procedures, undifferentiated complaint triage, and goals-of-care conversations — the judgment tier. Rises further if malpractice insurers require documented independent clinical reasoning separate from AI output.
Patient-facing rules requiring disclosure when an AI generated a clinical recommendation (Utah HB 452 style disclosure laws for mental health chatbots, extended to general care), plus payer requirements for an in-person examination before certain services — making a human encounter the billable, purchasable unit.
The limit. Already 77 and physically anchored; realistic headroom is a few points, mostly from practice-authority statutes. The main downside risk is not automation of the visit but scope substitution — if AI-assisted triage lets health systems staff more patients per clinician, headcount growth slows without any score changing.
| New York-Newark-Jersey City, NY-NJ | 16,800 | $167,650 +23% |
| Atlanta-Sandy Springs-Roswell, GA | 3,970 | $111,810 -18% |
| Los Angeles-Long Beach-Anaheim, CA | 3,890 | $161,890 +19% |
| Boston-Cambridge-Newton, MA-NH | 3,550 | $142,410 +5% |
| Philadelphia-Camden-Wilmington, PA-NJ-DE-MD | 3,390 | $135,640 +0% |
| Dallas-Fort Worth-Arlington, TX | 3,230 | $134,900 -1% |
| Chicago-Naperville-Elgin, IL-IN | 3,050 | $127,940 -6% |
| Phoenix-Mesa-Chandler, AZ | 3,050 | $134,650 -1% |
| San Jose-Sunnyvale-Santa Clara, CA | 1,030 | $223,640 +65% |
| Bakersfield-Delano, CA | 350 | $198,070 +46% |
| San Francisco-Oakland-Fremont, CA | 1,830 | $184,380 +36% |
We have no reported case of a named organisation automating this occupation. Not one deployment, not one announcement.
That is worth saying out loud next to a score of 77. The verdict above is about what the work exposes — what current AI could do to these tasks. It is not a claim that anyone has done it. For this occupation those two things have come apart completely: the capability argument is on this page, and the evidence column is empty.
Has AI actually changed your work? One tap, anonymous, and the running tally is public. Nothing else is asked of you.
Rather than check back: get the digest and we'll tell you what changed — or watch a single occupation from its own page.