Physician Assistants
29-1071.00Median wage $135,880/yr162,150 employed (US)Rank #699 of 923 scored · top 76% by substitution
Provide healthcare services typically performed by a physician, under the supervision of a physician. Conduct complete physicals, provide treatment, and counsel patients. May, in some cases, prescribe medication. Must graduate from an accredited educational program for physician assistants.
Sub-scores
0–100 · band = confidence interval from rater disagreement
Substitution19
Exposure21
Augmentation62
Substitution — the headline: capability discounted by cost, barriers and adoption.
Exposure — technical capability alone, regardless of whether anyone deploys it.
Augmentation — how much AI assists without replacing. High here + moderate substitution = a changing job, not a disappearing one.
Tasks on the substitution scale
12 rated tasks, binned by substitution score.
Position among all scored occupations
Distribution of 923 occupation scores; the marker is this occupation.
Tasks with substitution ≥ 70
0%
Run 1.0.0-draft.1 · computed 2026-08-05 · rater panel: claude-sonnet-5, claude-haiku-4-5-20251001 · intervals span rater disagreement.
Why this score
The five weighted dimensions of the composite, averaged across this occupation's tasks (importance-weighted, panel mean). Exact weights and formulas: /api/v1/methodology.
Task automatabilityw 35%18
panel mean rating 1.7/5 → substitution pressure 18/100
Technical feasibility todayw 20%25
panel mean rating 2.0/5 → substitution pressure 25/100
Cost vs. human wagew 15%21
panel mean rating 1.8/5 → substitution pressure 21/100
Adoption barriersw 20%inverted — strong barriers lower the score10
panel mean rating 4.6/5 (barrier strength) → substitution pressure 10/100
Sector adoption velocityw 10%26
panel mean rating 2.0/5 → substitution pressure 26/100
Task breakdown (12 tasks)
Substitution pressure per task, weighted by O*NET importance in the composite. Expand a task for the full rater audit trail — every rating, every model, every rationale.
Order medical and laboratory supplies and equipment.
59CI 45–72 · exposure 62 · augmentation 75 · importance 3.3/5 · click for rater detail
Order medical and laboratory supplies and equipment.
59| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 3/5 | Healthcare organizations have invested in EHR-integrated procurement systems and some use automated reordering for commodities, but adoption of agent-like autonomous ordering remains in pilot phases; most systems still require explicit human review and approval before fulfillment. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Healthcare supply chain digitization is progressing but unevenly, with many clinics still relying on manual or semi-manual ordering processes alongside more advanced hospital systems. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI can substantially augment PA ordering by providing real-time inventory visibility, suggesting optimal quantities based on historical use patterns, and flagging stock-outs or expired items—allowing the PA to complete ordering faster and more accurately while maintaining full oversight and decision authority. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI-driven inventory forecasting and automated reorder alerts significantly reduce the administrative burden on PAs, letting them focus on clinical duties while systems flag supply needs. |
| Task automatability | claude-haiku-4-5-20251001 | 3/5 | Ordering supplies can be partially automated through inventory management systems and AI-driven procurement workflows that identify needs and suggest orders, but requires human oversight for clinical appropriateness, budget constraints, and vendor selection—typically achieving moderate time savings rather than the 50%+ threshold for full automation. |
| Task automatability | claude-sonnet-5 | 4/5 | Ordering supplies based on inventory thresholds and standard catalogs is a structured, rule-based procurement task well within current automation capabilities via inventory management and e-procurement systems.' |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Healthcare supply ordering involves regulatory compliance (FDA, accreditation standards), liability for ordering errors affecting patient care, vendor contracts that require authorized personnel signatures, and organizational procurement policies that mandate human sign-off—creating substantial legal and operational friction. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No licensure requirement for ordering supplies, but organizational purchasing approval workflows, vendor contracts, and budget authorization create moderate friction. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 3/5 | AI-assisted procurement tools have comparable all-in costs to a PA's time on routine ordering (salary, system licensing, integration), with modest savings on repetitive inventory checks and ordering—roughly cost-neutral rather than an order-of-magnitude advantage. |
| Cost vs. human wage | claude-sonnet-5 | 4/5 | Automated procurement software is far cheaper per transaction than having a clinician manually track and order supplies, though integration and setup costs exist. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 3/5 | Procurement and inventory management systems exist and are widely deployed in healthcare, but they generally require human review before final ordering and do not autonomously handle exceptions, vendor negotiations, or complex supply-chain decisions at clinical-quality levels. |
| Technical feasibility today | claude-sonnet-5 | 4/5 | Automated inventory/reorder systems are widely deployed in hospitals and clinics for supply chain management, though physician assistants may still need to specify unusual or specialty items. |
Obtain, compile, and record patient medical data, including health history, progress notes, and results of physical examination.
37CI 32–41 · exposure 34 · augmentation 75 · importance 4.9/5 · click for rater detail
Obtain, compile, and record patient medical data, including health history, progress notes, and results of physical examination.
37| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 3/5 | Healthcare organizations are actively piloting EHR documentation tools and AI-assisted data capture, but adoption of full automation remains limited due to regulatory requirements and liability concerns; most deployments are augmentative rather than substitutional. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Healthcare has been steadily adopting ambient documentation tools in the last two years, though overall clinical workflow adoption remains uneven and pilot-heavy compared to fully digitized sectors. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI-powered transcription, data extraction, and documentation templates significantly assist PAs by reducing note-writing time and organizing patient data, allowing clinicians to focus on clinical reasoning while the system handles clerical structuring of information. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI scribes and dictation tools meaningfully reduce documentation burden and speed up note compilation, letting PAs focus more on patient interaction and exam. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can help extract and structure certain standardized data (lab results, vital signs), obtaining comprehensive medical history requires clinical judgment and patient rapport to identify relevant details. Compiling and recording into EHR systems is partially automatable but requires human verification for accuracy and legal liability, falling short of the 50% time-saving bar for end-to-end performance. |
| Task automatability | claude-sonnet-5 | 2/5 | AI ambient scribes can transcribe and structure encounter data, but obtaining history via patient interaction and performing physical exams requires human presence, so only the compilation/documentation portion is automatable.' |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Significant regulatory and liability barriers exist: a licensed PA must conduct the patient encounter and take responsibility for the accuracy and completeness of the medical record, which carries legal and malpractice implications. Automation cannot replace the legal requirement for a credentialed clinician to obtain and validate the history. |
| Adoption barriers | claude-sonnet-5 | 3/5 | Medical record accuracy and liability concerns mean PAs must review and attest to documentation, and data gathering via physical exam requires licensed personnel, creating moderate barriers. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | While transcription and data entry automation can reduce costs on specific subtasks, the end-to-end process still requires a PA's judgment and liability signing, meaning the full task cannot be performed cheaper than human labor when oversight costs are included. |
| Cost vs. human wage | claude-sonnet-5 | 3/5 | Scribe/documentation AI subscriptions cost far less than a human scribe, but the PA still must perform data gathering and examination themselves, limiting overall savings on the full task. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 3/5 | Deployed products exist for specific components (EHR documentation assistance, lab result parsing, transcription of clinical notes) but they operate within limited scope and require material human oversight. No system reliably performs the full task of obtaining and compiling patient data independently without error correction. |
| Technical feasibility today | claude-sonnet-5 | 3/5 | Ambient clinical documentation products (e.g., Nuance DAX, Abridge) are deployed in production and reliably generate structured notes from visits, though they still require clinician review and correction. |
Interpret diagnostic test results for deviations from normal.
32CI 28–36 · exposure 34 · augmentation 75 · importance 4.9/5 · click for rater detail
Interpret diagnostic test results for deviations from normal.
32| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 3/5 | Radiology and pathology AI adoption is accelerating in large health systems, but adoption remains uneven; many smaller practices and rural settings lag, and the fragmentation of EHR systems slows integration—suggesting middling, pilot-heavy adoption rather than deep production penetration. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Healthcare is adopting AI diagnostic aids steadily but unevenly, with pilots and narrow deployments (radiology, pathology) more advanced than broad diagnostic interpretation adoption across all test types. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI systems demonstrably assist PAs by rapidly highlighting abnormal values, suggesting reference ranges, and flagging critical findings, materially reducing review time and improving sensitivity; the human PA remains the decision-maker but gains significant productivity and diagnostic safety gains. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI decision-support tools meaningfully speed up screening, flagging, and triage of abnormal results, letting PAs focus attention and reduce oversight time while retaining final interpretive responsibility. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI systems can flag abnormal values and suggest patterns in lab/imaging data, interpreting results requires contextual clinical judgment, patient history integration, and nuanced reasoning about differential diagnoses—tasks that demand human oversight and cannot reliably achieve 50% time savings end-to-end today. |
| Task automatability | claude-sonnet-5 | 2/5 | AI can flag abnormal values and assist pattern recognition (e.g., imaging, labs) but full clinical interpretation requires integrating patient history and context, which current systems cannot reliably do end-to-end.calidad Full autonomous diagnostic interpretation is not yet at the 50% time-saving-at-equal-quality bar broadly across test types. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Strong regulatory and liability barriers exist: PAs bear legal responsibility for test interpretation, malpractice risk is high if AI errors cause harm, and medical boards/standards require human clinical judgment for diagnostic sign-off—effectively mandating human authority in the loop. |
| Adoption barriers | claude-sonnet-5 | 5/5 | Interpretation of diagnostic results for clinical decision-making legally requires a licensed provider's sign-off, with high liability exposure making AI substitution without human oversight essentially prohibited. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | Current AI interpretation tools (algorithms, platforms) have non-trivial licensing and integration costs, plus require expert human oversight to validate results; the all-in cost per test interpretation still exceeds a PA's incremental time cost for many routine assessments. |
| Cost vs. human wage | claude-sonnet-5 | 3/5 | Specialized AI diagnostic tools have meaningful licensing and integration costs comparable to marginal clinician time saved, though at scale for high-volume image-based tests cost per read can be lower. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 3/5 | AI products exist for specific test interpretation (e.g., radiology AI, pathology image analysis) with demonstrated accuracy in narrow domains, but production deployment remains inconsistent across healthcare systems, with material error rates and heavy reliance on human verification rather than autonomous decision-making. |
| Technical feasibility today | claude-sonnet-5 | 3/5 | FDA-cleared AI products exist for specific modalities (radiology, ECG, pathology) and are used in production, but coverage is narrow and most lab/test interpretation still relies on clinician judgment with AI as a secondary check. |
Instruct and counsel patients about prescribed therapeutic regimens, normal growth and development, family planning, emotional problems of daily living, and health maintenance.
27CI 25–29 · exposure 25 · augmentation 75 · importance 4.8/5 · click for rater detail
Instruct and counsel patients about prescribed therapeutic regimens, normal growth and development, family planning, emotional problems of daily living, and health maintenance.
27| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Healthcare remains a highly regulated, human-contact-required sector with slow AI adoption in clinical workflows. While telemedicine and chatbots are growing, actual displacement of PA counseling by autonomous systems is minimal and cautious. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Healthcare overall is a slower-adopting sector for patient-facing clinical judgment tasks, though administrative and documentation AI adoption is growing faster in adjacent areas. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI can meaningfully assist PAs by generating patient education summaries, providing evidence-based guidance on therapeutic regimens, and drafting counseling materials tailored to specific conditions. This augmentation improves productivity and consistency while the PA retains clinical judgment and patient interaction. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI tools can generate patient education handouts, summarize regimens in plain language, and help draft counseling talking points, meaningfully speeding up preparation while the PA still delivers personalized guidance. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can generate educational content and provide standardized information about therapeutic regimens or health maintenance, this task requires real-time clinical judgment, empathy, and adaptation to individual patient circumstances. Current systems cannot reliably counsel on complex emotional problems or family planning at equal quality to a PA, and oversight would be substantial. |
| Task automatability | claude-sonnet-5 | 2/5 | Personalized patient counseling requires building rapport, reading emotional cues, and adapting in real time to individual circumstances, which current AI cannot fully replicate despite being able to draft generic educational content. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Significant legal and regulatory barriers exist: a PA or physician must be responsible for clinical counseling in most jurisdictions, and liability for adverse outcomes from incorrect guidance creates asymmetric error costs. Patient preference for human interaction and informed-consent requirements add friction. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Counseling on prescribed regimens and health matters is part of licensed clinical scope of practice, with liability and regulatory expectations requiring a qualified provider to deliver or oversee this guidance. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 3/5 | AI-powered patient education platforms are relatively inexpensive to operate at scale, but require clinical oversight and integration into care workflows. The all-in cost including validation and human supervision is comparable to lower-cost PA time for routine counseling. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | While AI-generated content is cheap, the liability, need for clinician verification, and inability to fully replace the counseling interaction keep effective costs comparable to or higher than having a PA handle it directly. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | AI chatbots and educational tools exist for patient counseling, but they lack clinical licensing, cannot assess patient-specific risk factors, and often fail to handle nuanced emotional or sensitive topics appropriately. No deployed product reliably replaces a PA's counseling role in production healthcare settings. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Chatbots and patient portals provide some educational materials and FAQs, but no deployed product independently conducts clinical counseling on regimens, family planning, or emotional issues in production at scale. |
Administer or order diagnostic tests, such as x-ray, electrocardiogram, and laboratory tests.
24CI 20–28 · exposure 30 · augmentation 75 · importance 4.9/5 · click for rater detail
Administer or order diagnostic tests, such as x-ray, electrocardiogram, and laboratory tests.
24| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 3/5 | EHR-integrated clinical decision support is common in large healthcare systems, and many PAs use AI-assisted interpretation of images and labs. However, autonomous test ordering remains rare in production; adoption is primarily pilot-stage decision aids rather than replacement systems. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Healthcare adoption of AI for clinical ordering decisions remains slow due to regulatory, liability, and EHR integration hurdles, with pilots more common than full production deployment. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI substantially augments this task through diagnostic interpretation tools (radiology AI, lab analysis), order-set suggestions, and protocol reminders within EHRs, allowing PAs to order more efficiently and with greater confidence. The PA remains the decision-maker but AI materially raises their throughput and diagnostic accuracy. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI-driven clinical decision support and EHR order-entry systems increasingly help PAs identify appropriate tests faster and flag guideline-concordant options, meaningfully boosting efficiency while the PA retains decision authority. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can assist in interpreting diagnostic test results (image analysis, pattern recognition), the task of *administering or ordering* tests requires clinical judgment about patient presentation, contraindications, and cost-benefit reasoning that AI cannot yet perform independently. The ordering decision itself involves discretionary medical reasoning where AI cannot reliably substitute without human oversight. |
| Task automatability | claude-sonnet-5 | 2/5 | Ordering tests requires clinical judgment integrated with patient context, and physical administration (drawing blood, positioning for x-ray) requires hands-on physical action AI cannot perform; only the decision-support portion is partially automatable today. |
| Adoption barriers | claude-haiku-4-5-20251001 | 5/5 | Prescribing and ordering diagnostic tests is a licensed clinical function; PAs have legal authority to order tests within their scope, but autonomous AI ordering of medical tests would face substantial regulatory and liability barriers. A human clinician must typically review and sign orders for legal and safety reasons. |
| Adoption barriers | claude-sonnet-5 | 5/5 | Ordering diagnostic tests is a licensed clinical act requiring PA/physician authority, with legal and liability requirements mandating human sign-off and accountability for patient safety. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | Decision-support and interpretation tools are available but typically require PA licenses and oversight costs to deploy safely. The all-in cost (software, validation, compliance, PA time for review) remains comparable to or higher than direct PA ordering, especially given regulatory and malpractice risk. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | AI decision-support software has low marginal cost, but the actual administration of tests still requires human technicians and equipment, keeping overall cost comparable to current staffing models. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 3/5 | Products exist for specific subtasks: radiology AI for interpretation and order-set decision-support systems in EHRs can suggest appropriate tests. However, these tools require PA review and clinical input; no system reliably orders tests end-to-end without human validation, and liability concerns keep humans in the loop. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Clinical decision support tools exist to suggest appropriate diagnostic tests, but no deployed product autonomously orders or administers tests without PA/physician sign-off and physical execution by staff. |
Make tentative diagnoses and decisions about management and treatment of patients.
23CI 20–25 · exposure 30 · augmentation 75 · importance 4.9/5 · click for rater detail
Make tentative diagnoses and decisions about management and treatment of patients.
23| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Healthcare adoption of AI for autonomous diagnosis and treatment decisions remains limited and cautious, with most deployments serving as assistive tools rather than decision-makers. Regulatory uncertainty, liability concerns, and cultural resistance to algorithm-driven clinical autonomy slow deep adoption in this sector. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Healthcare is a historically slow-adopting sector for autonomous clinical decision-making due to regulation, liability, and safety concerns, though AI-assisted diagnostic tools are being piloted. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI augmentation is substantial and already demonstrable: differential diagnosis generators, clinical literature search, drug interactions, and evidence summaries meaningfully assist PAs in diagnosis and treatment planning. AI-powered decision support raises clinician productivity while the PA retains full responsibility and oversight. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI clinical decision support, differential diagnosis generators, and evidence retrieval tools meaningfully speed up and inform a PA's diagnostic reasoning and treatment planning while the human retains final judgment. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can assist with differential diagnosis generation and evidence lookup, end-to-end diagnostic and treatment decisions require integration of complex patient history, physical exam findings, and nuanced clinical judgment. Current AI systems cannot reliably achieve ≥50% time savings at equal quality across the full diagnostic and treatment-decision workflow without substantial human oversight. |
| Task automatability | claude-sonnet-5 | 2/5 | Diagnostic reasoning requires synthesizing physical exam findings, patient history, and clinical judgment under uncertainty and liability; current AI can suggest differentials but cannot reliably perform this end-to-end at equal quality without a licensed clinician making the actual decision.' |
| Adoption barriers | claude-haiku-4-5-20251001 | 5/5 | Strong legal and regulatory barriers exist: PAs must be licensed, supervised by or collaborating with physicians, and are directly liable for diagnostic and treatment decisions. Malpractice liability, state medical board regulations, and patient-safety standards create hard requirements for licensed human judgment and accountability. |
| Adoption barriers | claude-sonnet-5 | 5/5 | Diagnosis and treatment decisions are legally required to be made or authorized by licensed medical professionals, with significant malpractice liability attached, making this one of the most protected clinical tasks. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI diagnostic and treatment planning tools require significant infrastructure, maintenance, oversight, and liability buffering; inference costs plus integration and clinician review time remain comparable to or exceed a PA's loaded hourly cost for this high-stakes decision-making task. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | AI diagnostic support tools are cheap per query, but the requirement for a licensed PA/physician to review and finalize decisions means the human cost is still fully incurred, keeping overall cost comparable or higher when factoring liability and oversight. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 3/5 | Clinical decision support products (e.g., diagnostic assistants, drug interaction checkers) exist and perform narrowly in production, but no deployed system reliably makes independent tentative diagnoses and treatment decisions at the PA's level of clinical judgment. Error rates and scope limitations prevent full production deployment without physician supervision. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Clinical decision support tools and diagnostic AI exist (e.g., symptom checkers, imaging aids) but are used as adjuncts, not autonomous diagnosticians, and no deployed product independently makes treatment decisions in production. |
Prescribe therapy or medication with physician approval.
16CI 11–20 · exposure 17 · augmentation 75 · importance 4.9/5 · click for rater detail
Prescribe therapy or medication with physician approval.
16| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Healthcare is a high-regulation, liability-sensitive sector with slow AI adoption outside narrow niches (radiology reading assistance, administrative tasks). Prescription automation specifically is treated cautiously by health systems due to patient safety and legal exposure, resulting in limited real-world deployment. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Healthcare is a highly regulated, slower-adopting sector for autonomous clinical decisions, though AI-assisted diagnostic and prescribing support tools are gradually being piloted. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI excels at augmenting prescribing workflows: rapid drug-interaction checking, contraindication flagging, guideline-based recommendations, and evidence summaries meaningfully assist PAs in faster, safer decision-making while they remain accountable. This is one of the clearest augmentation wins in clinical informatics. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI tools meaningfully assist PAs by suggesting drug interactions, dosing guidance, and treatment options, improving speed and safety while the PA retains final decision authority. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can assist in medication selection through evidence-based recommendations and drug interaction checking, autonomous prescription—even with approval workflows—requires clinical judgment about individual patient contexts that current AI systems cannot reliably perform end-to-end. The physician approval requirement means AI would need to be more than a decision-support tool to achieve 50% time savings, which it cannot do at equivalent quality today. |
| Task automatability | claude-sonnet-5 | 1/5 | Prescribing requires clinical judgment, patient-specific risk assessment, and legal accountability that current AI cannot autonomously perform end-to-end.PA-level clinical decision-making with legal prescribing authority is not something off-the-shelf AI can replace today. |
| Adoption barriers | claude-haiku-4-5-20251001 | 5/5 | Prescribing is legally restricted to licensed healthcare providers (physicians, PAs, NPs) in nearly all jurisdictions; regulations explicitly require a human professional to take legal and clinical responsibility. Even with 'physician approval,' autonomous AI prescription would face direct regulatory prohibition and malpractice liability barriers. |
| Adoption barriers | claude-sonnet-5 | 5/5 | Prescribing is a legally regulated act requiring licensure, physician oversight, and DEA/state authority; no AI system can legally prescribe or approve medication independently. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | Current AI clinical decision-support tools require significant integration, validation, and oversight infrastructure, and they operate as assistants rather than replacements. The cost of maintaining compliant, auditable systems plus required human review likely exceeds savings from reduced PA time spent on routine medication selection. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | AI decision-support tools are cheap to run, but the requirement for licensed human sign-off means the human cost is not eliminated, keeping overall cost comparable to current practice. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | Decision-support products exist (e.g., drug-interaction checkers, guideline-based recommendations), but no deployed AI system can autonomously prescribe medication or generate clinically reliable prescriptions without substantial human oversight. Products in this space remain narrow-scope or research-stage; none handle the full complexity and liability of prescription generation reliably. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Clinical decision support tools exist and suggest medications or flag interactions, but no deployed product independently prescribes therapy with reliability accepted for autonomous clinical use. |
Examine patients to obtain information about their physical condition.
6CI 0–11 · exposure 8 · augmentation 50 · importance 4.9/5 · click for rater detail
Examine patients to obtain information about their physical condition.
6| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 1/5 | Healthcare remains a laggard sector for clinical task automation due to regulatory, liability, and human-contact requirements; physical examination by autonomous systems has seen minimal adoption in practice. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Healthcare adoption of AI is growing but concentrated in documentation and decision support, not physical examination itself, which remains untouched by automation. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can assist by summarizing vital signs, flagging abnormal lab patterns, or suggesting differential diagnoses based on patient data, moderately improving PA workflow and decision support during the examination process. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI can assist by suggesting differential diagnoses, flagging findings, or transcribing exam notes via ambient documentation tools, but it doesn't perform the hands-on examination. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | Physical examination requires direct patient contact, real-time clinical observation, and tactile assessment (palpation, auscultation, percussion) that current AI systems cannot perform autonomously. Remote AI might assist in analyzing images or vital signs post-collection, but cannot conduct the examination itself end-to-end. |
| Task automatability | claude-sonnet-5 | 1/5 | Physical examination requires hands-on palpation, auscultation, and direct sensory observation of a patient's body, which current AI cannot perform without embodiment; no off-the-shelf system can substitute end-to-end. |
| Adoption barriers | claude-haiku-4-5-20251001 | 5/5 | Legal and regulatory requirements mandate that a licensed provider (PA or physician) must perform or directly supervise patient examination; liability and standard-of-care expectations create hard barriers to full automation regardless of technical capability. |
| Adoption barriers | claude-sonnet-5 | 5/5 | Physical examination is a licensed clinical act requiring direct human-patient contact, with strong legal, regulatory, and liability requirements mandating a credentialed provider. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 1/5 | AI-assisted diagnostics are emerging, but the cost of infrastructure, oversight, liability, and integration remains high relative to the marginal savings on examination time, which represents only a portion of a PA's billable work. |
| Cost vs. human wage | claude-sonnet-5 | 1/5 | AI cannot perform the physical exam itself, so there is no viable cost comparison for substitution; a human clinician remains mandatory. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 1/5 | No deployed product performs autonomous patient physical examination. AI can analyze submitted test results or images, but no system can independently conduct the examination, take history, and synthesize findings in real clinical settings today. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product performs physical patient examinations; AI diagnostic tools only interpret data already collected by a human examiner. |
Supervise and coordinate activities of technicians and technical assistants.
6CI 0–11 · exposure 8 · augmentation 50 · importance 3.7/5 · click for rater detail
Supervise and coordinate activities of technicians and technical assistants.
6| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 1/5 | Healthcare organizations have shown minimal adoption of AI for team supervision; the requirement for human clinical authority, accountability, and interpersonal judgment makes this one of the slowest-moving automation frontiers in medicine. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Healthcare administrative workflows are adopting AI tools slowly for scheduling and documentation, but direct staff supervision remains untouched by automation trends. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI could provide useful assistance through scheduling optimization, performance analytics dashboards, or real-time alerts for workflow issues, but the physician assistant remains the essential decision-maker and authority in coordinating clinical technical staff. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI can assist with scheduling, task tracking, and communication logistics that support coordination, but the core supervisory judgment and interpersonal leadership remain human-driven. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | Supervision and coordination involve real-time human oversight, team management, and interpersonal decision-making that current AI cannot reliably handle autonomously. While AI could assist with scheduling or documentation, the core supervisory and coordinative functions require human authority and accountability. |
| Task automatability | claude-sonnet-5 | 1/5 | Supervising and coordinating human staff requires real-time interpersonal leadership, judgment about individual capabilities, and accountability that current AI cannot replicate end-to-end. |
| Adoption barriers | claude-haiku-4-5-20251001 | 5/5 | Supervision of technicians in a clinical setting is legally and practically tied to a licensed healthcare provider's authority, professional liability, and responsibility for patient outcomes—hard barriers that prevent substitution of the supervisory role. |
| Adoption barriers | claude-sonnet-5 | 5/5 | Supervisory responsibility in clinical settings typically requires a licensed provider to be accountable for staff actions and patient safety, a hard regulatory/liability barrier. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 1/5 | The cost of an AI system capable of reliable team supervision would far exceed the efficiency gains, especially given the oversight overhead and the need for a human supervisor to ultimately validate decisions and maintain accountability. |
| Cost vs. human wage | claude-sonnet-5 | 1/5 | There is no viable AI substitute performing this supervisory role, so cost comparison favors the human by default since AI cannot deliver the output. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 1/5 | No deployed AI product reliably performs direct supervision and coordination of technical teams in healthcare settings today; this requires licensed authority, real-time judgment, and legal accountability that remain firmly in human domain. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed AI product manages or supervises clinical technicians in production; scheduling/coordination tools exist but don't perform supervisory judgment or accountability functions. |
Provide physicians with assistance during surgery or complicated medical procedures.
4CI 0–7 · exposure 5 · augmentation 38 · importance 4.3/5 · click for rater detail
Provide physicians with assistance during surgery or complicated medical procedures.
4| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | While surgical robots are deployed in some high-volume specialties (urology, gynecology), adoption remains concentrated in large academic and specialty centers, and their use is for procedure execution under surgeon control, not autonomous PA replacement. Penetration is slow outside elite institutions. |
| Sector adoption velocity | claude-sonnet-5 | 1/5 | Surgical and procedural settings show minimal adoption of autonomous AI for physical assistance tasks, remaining a highly analog, hands-on domain. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | Surgical robots and visualization systems can assist surgeons in specific tasks (imaging guidance, precise instrumentation), but the core PA role—anticipating surgeon needs, managing the field, adjusting retractors, handing instruments—remains largely human-dependent; some augmentation of perception and precision is possible but gains are limited. |
| Augmentation potential | claude-sonnet-5 | 2/5 | AI can support pre-op planning, imaging analysis, or documentation, but offers little direct augmentation during the hands-on act of assisting in surgery itself. |
| Task automatability | claude-haiku-4-5-20251001 | 1/5 | Surgical assistance requires real-time physical manipulation, positioning, and decision-making in a sterile field alongside a surgeon. Current AI systems cannot perform surgical tasks end-to-end or achieve 50% time savings on the core assistance work; robotic surgical systems exist but require human control and are not autonomous assistants. |
| Task automatability | claude-sonnet-5 | 1/5 | Physical, hands-on intraoperative assistance requiring dexterity, sterile technique, and real-time judgment cannot be performed end-to-end by current AI systems. |
| Adoption barriers | claude-haiku-4-5-20251001 | 5/5 | Surgical assistance is legally and professionally regulated; a licensed physician assistant or nurse must perform intraoperative assistance, and liability for complications is borne by qualified humans. There is a strict legal requirement that trained, credentialed personnel provide surgical support. |
| Adoption barriers | claude-sonnet-5 | 5/5 | Surgical assistance is strictly regulated, requires licensure and hands-on physical presence, and carries high liability, making replacement legally and practically infeasible. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 1/5 | Robotic surgical systems cost millions in capital expenditure plus ongoing maintenance, training, and integration, making per-procedure costs far exceed the wage of a surgical PA. The ROI is poor for the assistance function alone. |
| Cost vs. human wage | claude-sonnet-5 | 1/5 | There is no viable AI substitute for this physical task, so no cost comparison favors AI; a human PA remains necessary at any cost. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | Surgical robots can augment specific procedures under human control, but no deployed AI system autonomously provides surgical assistance as a replacement for a PA. Existing robotic platforms like da Vinci require surgeon operation rather than independent assistance. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product provides autonomous surgical assistance; robotic surgery systems are human-operated tools, not autonomous assistants. |
Visit and observe patients on hospital rounds or house calls, updating charts, ordering therapy, and reporting back to physician.
1CI 0–3 · exposure 0 · augmentation 50 · importance 4.5/5 · click for rater detail
Visit and observe patients on hospital rounds or house calls, updating charts, ordering therapy, and reporting back to physician.
1| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 1/5 | Given the legal and safety barriers, healthcare organizations cannot and are not adopting AI to replace the human clinician presence required during rounds or house calls, even in highly digitized hospital settings. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Healthcare delivery, especially bedside/in-person care, is a slower-adopting sector for AI automation despite rapid AI uptake in administrative and diagnostic support functions. |
| Augmentation potential | claude-haiku-4-5-20251001 | 2/5 | AI can assist by auto-populating chart fields, suggesting relevant therapies, or flagging abnormal results, but the core task—visiting, observing, and assessing patients—remains entirely human-driven; augmentation is limited to peripheral documentation and note-taking support. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can meaningfully assist with chart updating (e.g., ambient documentation, voice-to-text note generation), order suggestions, and summarizing findings for physician reporting, even though the physical visit itself is unaided. |
| Task automatability | claude-haiku-4-5-20251001 | 1/5 | This task requires direct patient observation, physical examination, real-time clinical decision-making, and human judgment in a complex medical context. Current AI cannot perform bedside visits, observe subtle clinical signs, or make the dynamic clinical assessments necessary for this work. |
| Task automatability | claude-sonnet-5 | 1/5 | This task requires physical presence, hands-on observation, physical examination, and real-time clinical judgment in dynamic patient settings—none of which current AI systems can perform end-to-end. |
| Adoption barriers | claude-haiku-4-5-20251001 | 5/5 | Strong legal and regulatory barriers exist: state medical boards require physician assistants (or physicians) to be licensed practitioners who are present and responsible for patient care decisions. Liability, malpractice exposure, and patient safety regulations all mandate human clinical professionals in the loop. |
| Adoption barriers | claude-sonnet-5 | 5/5 | Direct patient examination and clinical decision-making by a licensed PA is legally required; scope-of-practice and licensure laws mandate human performance and physician oversight for this task. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 1/5 | AI systems cannot yet perform the core clinical functions of this task (patient observation, physical assessment, real-time clinical reasoning), so the cost comparison is moot; automation is not feasible at any price point today. |
| Cost vs. human wage | claude-sonnet-5 | 1/5 | There is no AI substitute performing the physical visit and examination component, so cost comparison favors the human by default since AI cannot deliver the core output. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 1/5 | No deployed AI system can independently conduct hospital rounds or house calls, observe patients in person, perform physical examinations, or make real-time clinical decisions that would replace a physician assistant's presence and judgment at the bedside. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product conducts patient rounds or house calls autonomously; this remains squarely within human-only clinical practice today. |
Perform therapeutic procedures, such as injections, immunizations, suturing and wound care, and infection management.
0CI 0–0 · exposure 0 · augmentation 25 · importance 4.5/5 · click for rater detail
Perform therapeutic procedures, such as injections, immunizations, suturing and wound care, and infection management.
0| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 1/5 | There is no measurable adoption of autonomous AI for performing these invasive medical procedures in clinical practice. Adoption remains near-zero because the technical, regulatory, and liability barriers are insurmountable with current technology. |
| Sector adoption velocity | claude-sonnet-5 | 1/5 | Physical medical procedures in clinical care settings show essentially no AI-driven automation or displacement; adoption is confined to administrative and diagnostic support, not hands-on procedures. |
| Augmentation potential | claude-haiku-4-5-20251001 | 2/5 | AI may assist PAs by providing diagnostic suggestions, procedure guidelines, or post-procedure monitoring support, but such augmentation is narrow and peripheral to the core manual execution of the procedure itself, which remains entirely human-centered. |
| Augmentation potential | claude-sonnet-5 | 2/5 | AI can assist with documentation, protocol reference, or decision support around infection management, but offers little direct assistance during the physical execution of injections, suturing, or wound care. |
| Task automatability | claude-haiku-4-5-20251001 | 1/5 | These therapeutic procedures require fine motor control, real-time tactile feedback, and physical manipulation of medical instruments on living patients—capabilities that current AI systems fundamentally lack. While AI can assist in planning or guidance, the actual execution of injections, suturing, and wound care remains exclusively within human-operator domain today. |
| Task automatability | claude-sonnet-5 | 1/5 | This is a hands-on physical procedure requiring manual dexterity, sterile technique, and real-time tactile judgment that current AI systems cannot perform without embodiment in capable robotics, which do not exist for this purpose. |
| Adoption barriers | claude-haiku-4-5-20251001 | 5/5 | These procedures are legally restricted to licensed healthcare professionals (PAs, physicians, nurses) and carry direct liability for medical harm. Regulatory bodies (FDA, state medical boards) explicitly govern who may perform invasive procedures, and patients have legitimate expectations of human clinical judgment and accountability. |
| Adoption barriers | claude-sonnet-5 | 5/5 | These are licensed clinical procedures requiring certified medical training, with strict liability, scope-of-practice regulation, and legal requirements for a credentialed human provider. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 1/5 | The hardware, integration, regulatory validation, and human oversight costs for any hypothetical autonomous medical procedure system would vastly exceed the cost of a trained PA performing these tasks, and no such system exists in practice today. |
| Cost vs. human wage | claude-sonnet-5 | 1/5 | There is no AI substitute performing these physical procedures, so no meaningful cost comparison exists; the human is the only viable option. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 1/5 | No deployed AI product reliably performs these physical medical procedures autonomously or at scale. Surgical robots exist for highly constrained, pre-planned procedures in controlled settings, but they require human surgeons at the controls and are not autonomous systems addressing this task statement. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product performs injections, suturing, or wound care autonomously in clinical practice; this remains far outside current commercial AI capability. |
Related occupations — Healthcare Practitioners & Technical
How to read this
A high substitution score does not mean this job disappears — it means a large share of its current tasks face replacement pressure, so the mix of tasks is likely to change. High augmentation alongside substitution typically means the occupation reorganizes around the protected tasks. Wide confidence intervals mean the rater panel disagreed: treat those scores as open questions, not verdicts.
What would change this score
New model capabilities (automatability, feasibility), falling inference costs (cost ratio), regulation and licensing shifts (barriers), and measured sector adoption (velocity) all re-enter at every index release. Each release is recomputed, versioned and kept queryable — scores are claims with a date on them, not permanent labels.