Physical Therapists
29-1123.00Median wage $102,760/yr267,330 employed (US)Rank #752 of 923 scored · top 81% by substitution
Assess, plan, organize, and participate in rehabilitative programs that improve mobility, relieve pain, increase strength, and improve or correct disabling conditions resulting from disease or injury.
Sub-scores
0–100 · band = confidence interval from rater disagreement
Substitution17
Exposure15
Augmentation51
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
24 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%14
panel mean rating 1.6/5 → substitution pressure 14/100
Technical feasibility todayw 20%17
panel mean rating 1.7/5 → substitution pressure 17/100
Cost vs. human wagew 15%19
panel mean rating 1.8/5 → substitution pressure 19/100
Adoption barriersw 20%inverted — strong barriers lower the score18
panel mean rating 4.3/5 (barrier strength) → substitution pressure 18/100
Sector adoption velocityw 10%19
panel mean rating 1.7/5 → substitution pressure 19/100
Task breakdown (24 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.
Provide educational information about physical therapy or physical therapists, injury prevention, ergonomics, or ways to promote health.
55CI 51–59 · exposure 50 · augmentation 75 · importance 4.4/5 · click for rater detail
Provide educational information about physical therapy or physical therapists, injury prevention, ergonomics, or ways to promote health.
55| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 3/5 | Healthcare and physical therapy have moderate digitization and cautious but growing AI adoption. Clinics increasingly use templated or AI-assisted patient education, but adoption is slower than in information-heavy sectors due to regulatory caution, liability concerns, and the prevalence of small independent practices. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Healthcare, especially outpatient rehab settings, is a slower-adopting sector for patient-facing AI content generation compared to information/finance industries, though patient portals are slowly incorporating AI tools. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI can substantially assist PTs by rapidly drafting personalized education, generating visual explanations, and adapting content to patient literacy levels, allowing the PT to focus on clinical judgment and relationship-building rather than rote content creation. This is a natural augmentation use case with clear productivity gains. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can efficiently draft, translate, and customize educational materials on ergonomics and injury prevention, letting therapists spend more time on hands-on care while still reviewing and personalizing content. |
| Task automatability | claude-haiku-4-5-20251001 | 3/5 | AI can generate educational content about physical therapy, injury prevention, and ergonomics at scale, but creating personalized, context-sensitive education for diverse patient populations requires human judgment about individual needs, literacy levels, and cultural factors. Current systems can draft and deliver ~50% of this task reliably, but the therapeutic relationship and verification of comprehension often require human oversight. |
| Task automatability | claude-sonnet-5 | 3/5 | Generating educational content about injury prevention, ergonomics, and health promotion is well within current LLM capabilities, though personalizing it to a specific patient's condition and delivering it with rapport requires human judgment. |
| Adoption barriers | claude-haiku-4-5-20251001 | 3/5 | Regulatory barriers exist (PTs must ensure accuracy and appropriateness of patient education), and there is professional expectation that PTs verify educational content before distribution. However, these are oversight requirements rather than absolute licensing prohibitions, creating friction but not a legal barrier to automation. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No strict licensing requirement mandates a PT deliver general educational information, though clinical context and liability considerations create some preference for professional involvement. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 4/5 | AI-generated educational materials have minimal marginal cost per delivery once created, while a PT's time for patient education is hourly labor. At scale, AI content delivery is substantially cheaper per patient educated, though initial curation and PT validation add overhead. |
| Cost vs. human wage | claude-sonnet-5 | 4/5 | Generating written or verbal educational content via AI is extremely cheap compared to a therapist's billed time, even accounting for review and customization. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 3/5 | Educational chatbots and AI-generated patient materials exist in deployed healthcare settings, but they typically serve as supplementary resources rather than primary educators. Error rates in medical accuracy and the need for PT review before patient delivery mean these systems are narrowly scoped and require significant human gatekeeping. |
| Technical feasibility today | claude-sonnet-5 | 3/5 | Patient education chatbots and AI-generated handout tools exist and are used in some clinics, but they are not yet the standard reliable delivery mechanism for individualized clinical education across the profession. |
Record prognosis, treatment, response, and progress in patient's chart or enter information into computer.
50CI 39–61 · exposure 42 · augmentation 75 · importance 4.7/5 · click for rater detail
Record prognosis, treatment, response, and progress in patient's chart or enter information into computer.
50| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 4/5 | Healthcare IT adoption is accelerating, and speech-to-text EHR integration is becoming standard in many large clinic networks and hospital systems. Physical therapy practices increasingly use cloud-based documentation platforms compatible with AI-assisted inputs. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Healthcare, especially outpatient rehab therapy, has been slower than finance or professional services to adopt AI documentation tools, though EHR-integrated scribes are gaining some traction. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | Speech-to-text, AI-suggested templates, and automated extraction of vitals/ROM data can substantially reduce documentation time while the therapist reviews and finalizes the clinical narrative. This is a high-impact augmentation use case already deployed in many settings. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI-assisted note drafting and templated documentation meaningfully speeds up charting while the therapist remains responsible for accuracy and final entry. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | Recording structured data (treatment type, numeric progress metrics) could be partially automated from clinical notes or EHR signals, but prognosis interpretation and nuanced progress assessment require clinician judgment. Current AI cannot reliably capture the clinical reasoning needed for complete charting without significant human correction. |
| Task automatability | claude-sonnet-5 | 3/5 | AI dictation and ambient scribe tools can draft clinical notes from clinician speech or observations, but a human must still verify accuracy and finalize the record, so only partial time savings are realized end-to-end. |
| Adoption barriers | claude-haiku-4-5-20251001 | 2/5 | Documentation is a regulatory and liability requirement that the therapist must sign off on; however, there is no licensing barrier preventing AI pre-population or draft generation. Oversight and legal accountability remain with the clinician, creating modest friction but not a hard prohibition. |
| Adoption barriers | claude-sonnet-5 | 3/5 | Clinical documentation must meet legal, billing, and licensure standards requiring therapist review and sign-off, creating moderate but not absolute barriers to full automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 4/5 | Voice-to-text and template-filling tools have low inference costs relative to clinician time (~$50–100/hour loaded wage). Even accounting for oversight, AI-assisted charting is substantially cheaper than unaugmented manual documentation. |
| Cost vs. human wage | claude-sonnet-5 | 3/5 | Documentation software subscriptions plus review time are cheaper than pure manual charting but not dramatically so once integration, correction, and compliance checks are factored in. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 3/5 | Speech-to-text and basic EHR templating exist in production, but AI systems that automatically populate prognosis and treatment response fields with clinical accuracy are limited and typically require substantial physician oversight. Narrow products for structured data entry work, but end-to-end charting automation is not reliably deployed at scale. |
| Technical feasibility today | claude-sonnet-5 | 3/5 | Ambient clinical documentation products (e.g., speech-to-text/EHR scribes) are deployed in some healthcare settings including PT/rehab, but adoption in physical therapy specifically is narrower and error correction is still routine. |
Construct, maintain, or repair medical supportive devices.
41CI 5–76 · exposure 45 · augmentation 50 · importance 3.8/5 · click for rater detail
Construct, maintain, or repair medical supportive devices.
41| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 3/5 | Orthotic and prosthetic labs are piloting and deploying 3D printing and robotic manufacturing, but adoption remains patchy outside large hospital systems and specialized centers; small clinics and individual practices still rely heavily on manual methods. |
| Sector adoption velocity | claude-sonnet-5 | 1/5 | Physical device fabrication and repair in clinical settings show minimal AI adoption, as this is a hands-on, low-digitization task. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI-driven CAD generation, material optimization, and automated quality inspection significantly assist physical therapists in designing and validating custom devices, speeding iteration and reducing manual rework while the therapist retains final clinical judgment. |
| Augmentation potential | claude-sonnet-5 | 2/5 | AI/CAD tools and 3D printing design software can assist in designing or modeling supportive devices, but the construction, fitting, and repair remain largely manual. |
| Task automatability | claude-haiku-4-5-20251001 | 5/5 | Robotic arms and CNC systems can now design, fabricate, and assemble custom orthoses, braces, and prosthetics from digital models with high precision and speed, achieving well over 50% time savings compared to manual construction while maintaining or exceeding quality standards. |
| Task automatability | claude-sonnet-5 | 1/5 | This requires physical fabrication, fitting, and hands-on adjustment of orthotics, splints, or braces to a patient's body, which current AI systems cannot perform end-to-end." |
| Adoption barriers | claude-haiku-4-5-20251001 | 3/5 | Physical therapists must sign off on fit and medical suitability, and some liability attaches to device design; however, fabrication itself (construction, maintenance, repair) is increasingly decoupled from licensure and can be delegated to technicians or automated systems without regulatory override. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Fitting and adjusting medical devices to patients typically requires licensed clinical judgment and hands-on skill, with liability concerns around improper fit causing injury. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 4/5 | Additive manufacturing and robotic fabrication have dramatically reduced per-unit material and labor costs; amortized over high-volume production, AI-driven design and robotic construction costs are typically 50–80% cheaper than skilled manual labor for standard devices. |
| Cost vs. human wage | claude-sonnet-5 | 1/5 | There is no AI substitute performing this physical labor, so AI cost is not comparable or lower than human cost for this task. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 4/5 | Commercial 3D printing and robotic fabrication systems are actively deployed in orthotic/prosthetic labs and medical device manufacturing; while assembly complexity sometimes requires human oversight, the core construction and maintenance workflows are reliably automated in production settings. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed AI product constructs or repairs physical supportive devices; this remains a manual, craft-based clinical task. |
Conduct or support research and apply research findings to practice.
32CI 25–39 · exposure 30 · augmentation 75 · importance 3.5/5 · click for rater detail
Conduct or support research and apply research findings to practice.
32| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Adoption of AI-assisted research tools in physical therapy is slow and limited to larger academic institutions and research-focused clinics. Most physical therapy practices focus on direct patient care, not research; adoption of research automation in the profession lags far behind information-sector adoption patterns. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Healthcare and clinical research sectors have been slower to adopt AI deeply into practice compared to information/finance sectors, with pilots more common than production-scale use for research application tasks. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI substantially assists researchers by accelerating literature review, organizing references, performing statistical analyses, and drafting research summaries. These tools meaningfully raise the productivity of physical therapists conducting or applying research, even though humans retain full responsibility for interpretation and design decisions. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI tools significantly enhance literature review, evidence synthesis, and identification of relevant research findings, meaningfully boosting a PT's efficiency in staying current and applying evidence-based practice. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can assist with literature review, data analysis, and summarizing research findings, conducting research itself requires human judgment, experimental design, protocol ethics approval, and clinical interpretation. The task has a research *conduct* component that requires domain expertise and ethical oversight that AI cannot independently manage today. |
| Task automatability | claude-sonnet-5 | 2/5 | AI can assist with literature search, summarization, and statistical analysis, but designing studies, collecting clinical data, and translating findings into individualized patient care requires human clinical judgment that current systems cannot replicate end-to-end. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Research involving human subjects requires Institutional Review Board approval, informed consent, and licensed professional oversight of methodology and ethics. Licensing requirements, liability for study design flaws, and regulatory frameworks for human-subjects research create substantial legal and professional barriers to full automation. |
| Adoption barriers | claude-sonnet-5 | 3/5 | No strict licensing requirement mandates a human conduct research, but clinical application to practice involves professional judgment and institutional review processes that create moderate friction. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI tools (subscriptions to research databases, statistical software, or LLMs) cost money but do not replace the substantial time investment of a licensed physical therapist conducting research. The human still does most of the intellectual and coordination work, making all-in cost comparable to or higher than the human wage for research hours. |
| Cost vs. human wage | claude-sonnet-5 | 3/5 | AI tools can cut costs for literature synthesis and data analysis portions, but the overall task still requires substantial paid clinical expertise and oversight, keeping costs roughly comparable to human-only workflows for full task completion. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 3/5 | Products exist for research assistance—literature databases, statistical software, and AI-powered search tools—but none reliably conduct or validate research end-to-end. Applied research in clinical settings still requires human researchers for protocol design, patient recruitment, outcome measurement, and institutional review board interactions. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | AI literature review and research assistant tools exist and are used in academic settings, but no deployed product autonomously conducts or applies clinical PT research to practice reliably. |
Review physician's referral and patient's medical records to help determine diagnosis and physical therapy treatment required.
27CI 25–29 · exposure 25 · augmentation 63 · importance 4.6/5 · click for rater detail
Review physician's referral and patient's medical records to help determine diagnosis and physical therapy treatment required.
27| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Healthcare adoption of AI for clinical decision support remains slow outside radiology and pathology; physical therapy clinics are largely small to mid-sized private practices with lower digitization and AI integration compared to hospital systems or tech-forward sectors. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Healthcare, especially allied health like physical therapy, has historically lagged in AI adoption due to regulatory, liability, and workflow integration challenges relative to finance or tech sectors. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can meaningfully assist by highlighting key findings, extracting structured data from unstructured records, and summarizing relevant history, allowing the PT to focus on synthesizing the clinical picture rather than document review. This is useful but not transformative—the clinical reasoning remains human-driven. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can efficiently extract and summarize key information from physician referrals and medical records, saving significant time for PTs during the review phase even though final clinical judgment remains human-led. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can extract and summarize information from medical records and referrals, the task requires clinical judgment to synthesize complex patient history, imaging, and physician notes into a differential diagnosis and treatment plan. Current AI cannot reliably perform this interpretive synthesis at the quality and accountability level required in physical therapy. |
| Task automatability | claude-sonnet-5 | 2/5 | AI can summarize records and flag relevant history, but synthesizing this into a diagnosis and treatment plan requires clinical judgment and hands-on assessment that current systems cannot fully replace.4B |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Physical therapists are licensed professionals whose scope includes diagnosis and treatment planning; liability, scope-of-practice regulations, and organizational standards typically require a licensed PT to review records and make clinical judgments. Medical-legal accountability for misdiagnosis or inappropriate referral creates strong barriers to full automation. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Diagnosis and treatment planning in physical therapy are subject to licensure requirements and legal accountability, meaning a licensed PT must review and sign off, creating a strong regulatory barrier. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | Current AI document analysis and summarization tools are relatively inexpensive, but the human PT oversight required to validate diagnostic reasoning and treatment plans means the all-in cost still approaches or exceeds the hourly rate for initial evaluation work. |
| Cost vs. human wage | claude-sonnet-5 | 3/5 | AI-assisted summarization is cheap per record, but human PT review and liability oversight remain necessary, keeping overall cost comparable to human-only workflows once integration and correction are counted. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | AI tools can assist with document review and information extraction, but no deployed product reliably performs the full diagnostic and treatment-planning synthesis that a PT must do. Products exist for note summarization and coding, but clinical decision-making at this level remains research-stage or narrow in scope. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Clinical documentation summarization tools exist and are used in some EHR workflows, but no deployed product reliably performs full diagnostic-relevant chart review and treatment determination for PT autonomously. |
Identify and document goals, anticipated progress, and plans for reevaluation.
25CI 25–25 · exposure 25 · augmentation 50 · importance 4.6/5 · click for rater detail
Identify and document goals, anticipated progress, and plans for reevaluation.
25| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Adoption of autonomous or near-autonomous AI for this task in healthcare is slow; most PT practices remain traditional in goal documentation, with limited pilot adoption of AI-driven planning systems. The sector is moderately digitized but resistant to ceding clinical judgment to automation. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Healthcare/outpatient rehab settings are relatively slow AI adopters for clinical decision content, though documentation-assist tools are gradually appearing. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can usefully assist by drafting goal templates, suggesting evidence-based timelines, and flagging prior functional trends from patient records, thus reducing documentation burden and improving consistency. However, the therapist must still validate and modify goals based on clinical judgment and patient input. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI-based scribes and templated note generators can help therapists draft and organize goal statements and reevaluation timelines faster, while the therapist retains clinical decision-making. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can generate draft text for goals and reevaluation timelines based on templates and patient data, the task fundamentally requires clinical judgment about patient-specific functional outcomes, prognosis, and individualized treatment trajectories. Current AI systems cannot reliably synthesize complex patient histories, comorbidities, and functional status into clinically sound, patient-centered goal-setting at the quality level required for therapeutic accountability. |
| Task automatability | claude-sonnet-5 | 2/5 | Drafting goal templates and boilerplate can be automated, but setting clinically appropriate goals and reevaluation plans requires patient-specific clinical judgment that AI cannot reliably perform end-to-end today. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Strong barriers exist: a licensed physical therapist must legally establish, document, and sign-off on treatment goals and reevaluation plans, and clinical liability for inappropriate goal-setting is substantial. Regulatory oversight (state licensure, Medicare conditions of participation) and professional standards require human clinical ownership of this planning task. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Licensed PTs are legally required to establish and sign off on treatment goals and plans of care, and payers/regulators require clinician authorship and accountability for these documents. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | Integration costs for specialized EHR modules and the need for human clinical review and sign-off mean the all-in cost per fully-executed task remains comparable to or exceeds direct therapist time, especially given liability and accuracy requirements. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | AI can cheaply generate draft text, but the clinician must still review, verify, and finalize goals against exam findings, limiting overall cost savings versus the licensed therapist's time. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | No deployed product reliably performs independent goal-setting and reevaluation planning in physical therapy. Some EHR systems offer templated prompts and structured documentation aids, but these still require therapist review and clinical decision-making; they are assistive tools, not autonomous performers of the task. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Some clinical documentation and EHR templating tools assist with drafting goal language, but no deployed product independently sets valid PT treatment goals and reevaluation plans in production. |
Instruct patient and family in treatment procedures to be continued at home.
23CI 16–30 · exposure 17 · augmentation 75 · importance 4.7/5 · click for rater detail
Instruct patient and family in treatment procedures to be continued at home.
23| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Physical therapy remains a hands-on, human-contact service with limited digitization; adoption of AI for clinical instruction is pilot-stage at best, concentrated in research settings rather than widespread production use in clinics. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Healthcare/rehab settings are generally slower adopters of AI for direct patient interaction tasks compared to information-sector benchmarks, though patient education content generation is beginning to see pilot use. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI can meaningfully assist by generating illustrated exercise guides, creating personalized video demonstrations, or organizing instructional materials—allowing therapists to spend more time on assessment and real-time feedback rather than manual documentation and basic content creation. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can meaningfully assist therapists by generating customized exercise handouts, videos, and reminders that the therapist then reviews and delivers, improving efficiency while the therapist remains central to instruction. |
| Task automatability | claude-haiku-4-5-20251001 | 1/5 | Instructing patients and families in home treatment requires real-time personalized assessment, adaptive explanation based on patient comprehension and physical capability, and human judgment about readiness—capabilities that current AI systems cannot reliably deliver end-to-end in clinical settings. |
| Task automatability | claude-sonnet-5 | 2/5 | AI can generate written or video home-exercise instructions, but personalized instruction based on live assessment of a patient's condition, adherence barriers, and family dynamics still requires human judgment and interaction.4/5 not met. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Physical therapists are licensed professionals whose scope explicitly includes patient education and safety assessment; liability and regulatory requirements strongly favor human sign-off on home therapy instructions, particularly for injury or post-operative populations where errors carry clinical risk. |
| Adoption barriers | claude-sonnet-5 | 3/5 | No strict licensing requirement mandates the therapist perform this specific instructional step, but liability, patient safety, and the need for hands-on demonstration create moderate friction against full automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI-generated instructional content is cheap, but the task's core value lies in expert assessment and personalized explanation that currently requires a licensed therapist; substitution would save only the content-delivery overhead, not the bulk of the labor cost. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | AI-generated handouts/videos are cheap to produce, but the overall task still requires therapist time for assessment, demonstration, and Q&A, so cost savings are only partial. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | While AI can generate instructional videos or written exercise guides, no deployed system reliably assesses individual patient needs, adjusts instruction for comprehension, or takes responsibility for safe home execution; products exist for content delivery but not for the clinical instruction role itself. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Some patient education apps and AI-generated home exercise plans exist, but they are supplementary tools rather than reliable substitutes for a therapist's tailored instruction in production settings. |
Refer clients to community resources or services.
23CI 16–30 · exposure 17 · augmentation 50 · importance 3.8/5 · click for rater detail
Refer clients to community resources or services.
23| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Healthcare and PT practices remain relatively slow in automation adoption; while some practices use basic resource directories, active AI-driven referral systems are rare and adoption is limited by regulatory caution and human-centered care norms. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Healthcare, especially outpatient rehabilitation settings, has historically been slow to adopt AI-driven administrative and care-coordination tools compared to finance or tech sectors. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI could usefully assist by surfacing and summarizing available community resources, allowing therapists to review options more quickly; however, the core matching and judgment remain human responsibilities. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI can help therapists identify relevant local resources, generate referral letters, and keep track of community services, offering moderate productivity benefits while the therapist retains final judgment. |
| Task automatability | claude-haiku-4-5-20251001 | 1/5 | Referring clients to resources requires understanding individual circumstances, needs assessment, and judgment about fit—tasks requiring human discretion and contextual knowledge that current AI cannot reliably perform end-to-end without human oversight. |
| Task automatability | claude-sonnet-5 | 2/5 | Identifying and referring clients to community resources requires clinical judgment about the patient's evolving needs, which AI can support but not fully replace end-to-end today.4 |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Physical therapists are licensed professionals whose clinical judgment and legal responsibility for client welfare create strong barriers; inappropriate referrals can harm clients, and liability rests with the clinician who must personally verify and endorse recommendations. |
| Adoption barriers | claude-sonnet-5 | 3/5 | While not strictly requiring a license to give a referral, clinical accountability, patient trust, and liability for inappropriate referrals create moderate friction against full automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI systems for resource matching have non-trivial setup, maintenance, and oversight costs, and still require human review to ensure safety and appropriateness—making them comparably expensive or more expensive than direct human referral. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | Human clinicians already do this quickly as part of routine care, so AI tools add licensing/integration costs without dramatically undercutting the marginal cost of a brief referral conversation. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | While AI could retrieve and list available community resources, deployed systems lack the clinical judgment and personalized assessment needed to appropriately match clients to services; no mature product reliably does this task in production PT settings. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Some care-coordination software can suggest resources or generate referral lists, but no deployed product autonomously manages the referral decision and communication reliably in clinical practice. |
Perform and document an initial exam, evaluating data to identify problems and determine a diagnosis prior to intervention.
23CI 20–25 · exposure 25 · augmentation 63 · importance 4.8/5 · click for rater detail
Perform and document an initial exam, evaluating data to identify problems and determine a diagnosis prior to intervention.
23| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Healthcare adoption of AI remains moderate and cautious; diagnostic decisions in physical therapy are still predominantly human-led despite digital documentation systems, reflecting regulatory and professional conservatism. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Healthcare, especially hands-on rehabilitative care, adopts AI slowly relative to information-sector norms, with pilots for documentation but little diagnostic automation in practice. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can meaningfully assist with documentation templating, imaging analysis, and guideline-based decision support during the exam, raising clinician efficiency without replacing the necessity of hands-on assessment and professional judgment. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI ambient documentation and clinical decision-support tools meaningfully speed up charting and can help organize evaluation data, improving therapist productivity while they remain in control of the exam and diagnosis. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | AI can assist in data interpretation (imaging, vital signs) but cannot independently perform physical examination (palpation, range-of-motion testing, patient interaction) or make clinical diagnoses that require direct patient assessment, judgment, and legal accountability. |
| Task automatability | claude-sonnet-5 | 2/5 | The hands-on physical examination and clinical judgment required cannot currently be performed end-to-end by AI; only the documentation/data-synthesis portion is automatable, well short of full task automation with equal quality. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Physical therapy diagnosis and treatment planning require state licensure and professional liability; only a licensed PT can legally perform initial examination and establish diagnosis, creating a hard legal barrier to full automation. |
| Adoption barriers | claude-sonnet-5 | 5/5 | Diagnosis and initial evaluation in physical therapy require a licensed practitioner by law and professional standards, creating a hard regulatory and liability barrier to substitution. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | Integration of AI for documentation and imaging review adds cost and overhead to a license-protected professional service; savings do not approach the loaded wage of a physical therapist for the complete diagnostic task. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | AI can cheaply assist with note-taking, but the exam itself still requires the licensed therapist's time, so overall cost savings are modest, not order-of-magnitude. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | Narrow-scope products exist for documentation and imaging interpretation, but no deployed system can reliably perform the full initial diagnostic exam independently; clinical judgment and physical examination remain essential human functions in production settings. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | AI scribe and documentation-assist tools exist but no deployed product performs the physical examination or generates a reliable diagnosis independently in production PT settings. |
Teach physical therapy students or those in other health professions.
21CI 16–25 · exposure 17 · augmentation 63 · importance 3.7/5 · click for rater detail
Teach physical therapy students or those in other health professions.
21| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Healthcare education has been slow to adopt AI-led instruction; most deployments remain pilot supplementary tools rather than replacements, and sectors like physical therapy education prioritize hands-on clinical training from licensed professionals. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Higher education and allied health training are moderate-to-slow adopters of AI for hands-on clinical instruction, with adoption concentrated in supplementary digital materials rather than core teaching. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can assist instructors by generating lecture notes, creating practice quizzes, and producing explanatory diagrams, but the instructor remains essential for live feedback, technique correction, and student assessment. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can meaningfully assist instructors by generating course materials, case studies, quizzes, and personalized study aids, augmenting teaching efficiency significantly. |
| Task automatability | claude-haiku-4-5-20251001 | 1/5 | Teaching health professions requires real-time interaction, assessment of student understanding, dynamic adjustment of explanations, and hands-on demonstration of physical techniques—capabilities that current AI systems cannot perform end-to-end at equal quality as experienced instructors. |
| Task automatability | claude-sonnet-5 | 2/5 | Teaching involves live demonstration, hands-on correction, and adaptive mentorship that current AI cannot replicate end-to-end, though AI can generate some lecture content or quizzes.' |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Educational accreditation standards, professional licensing bodies, and institutional policy typically mandate that licensed physical therapists or qualified faculty deliver core instruction; liability for inadequate student preparation creates strong incentives to retain human oversight. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Accredited PT education requires qualified faculty and clinical supervision under accreditation standards, creating strong institutional and regulatory barriers to full AI substitution. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI-generated supplementary materials are cheap, but replacing an instructor's salary with AI would require both content generation and live interactive systems; the integrated cost remains high relative to the task's value delivery, and most institutions still rely on human instructors. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | Human clinical instructors remain necessary for supervised practice and licensure-relevant skill certification, so AI substitution saves little of the true cost despite cheap content generation tools. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | While AI can generate educational content and tutoring systems exist, no production system reliably replaces the core teaching functions: live demonstration of complex physical movements, real-time observation and correction of student technique, and the interpersonal rapport essential to healthcare education. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | AI tutoring products exist for didactic content but no deployed system reliably teaches clinical/manual physical therapy skills to students in production settings. |
Test and measure patient's strength, motor development and function, sensory perception, functional capacity, or respiratory or circulatory efficiency and record data.
15CI 5–25 · exposure 17 · augmentation 50 · importance 4.6/5 · click for rater detail
Test and measure patient's strength, motor development and function, sensory perception, functional capacity, or respiratory or circulatory efficiency and record data.
15| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 1/5 | Healthcare, especially physical rehabilitation, is slower to adopt fully autonomous AI systems due to regulatory scrutiny, malpractice liability, and patient-contact requirements; adoption remains primarily in research and pilot phases. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Healthcare/physical rehabilitation is a slower-adopting sector for AI automation of hands-on clinical tasks, though digital documentation tools are spreading. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI-assisted tools (e.g., automated video analysis of gait, real-time sensor feedback during testing, data charting) can help PTs document and interpret measurements more efficiently, though the human PT must perform and validate the core assessments. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI-enabled sensors, wearables, and documentation tools can help quantify and record patient metrics, aiding the therapist's assessment and reducing manual charting time. |
| Task automatability | claude-haiku-4-5-20251001 | 1/5 | While some measurement data (e.g., range of motion angles, force gauge readings) could be digitally logged, the core of this task requires hands-on physical assessment—palpation, observation of movement quality, patient response to resistance—which current AI cannot perform without human execution of the actual tests. |
| Task automatability | claude-sonnet-5 | 2/5 | Physical measurement of strength, motor function, and sensory perception requires hands-on assessment and clinical judgment that current AI cannot perform end-to-end; AI can assist with recording/interpreting data but not conducting the physical exam. |
| Adoption barriers | claude-haiku-4-5-20251001 | 5/5 | State licensing laws explicitly require a licensed physical therapist to perform patient evaluation and assessment; automation of these gatekeeping clinical tasks faces both legal/regulatory barriers and liability concerns around autonomous medical measurement without professional oversight. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Physical therapy assessment is a licensed clinical task requiring hands-on contact and professional judgment, with liability and regulatory requirements mandating a licensed PT perform or supervise it. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 1/5 | The hardware and software needed to automate functional assessment (motion capture, sophisticated sensors, integration) plus clinical validation costs exceed what a PT charges per session, making the all-in cost higher than human labor. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | Sensor/software tools add cost on top of clinician time rather than replacing it, so AI-assisted assessment is not clearly cheaper than a therapist performing the exam directly. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | Some narrow components like automated ROM logging from video or accelerometer data exist in research/pilot settings, but no deployed product reliably replaces a PT's full battery of strength, motor, sensory, and functional tests. Clinical adoption remains minimal. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Some wearable sensors and motion-capture tools exist to quantify movement, but no deployed product autonomously performs full physical therapy assessments in production at scale. |
Plan, prepare, or carry out individually designed programs of physical treatment to maintain, improve, or restore physical functioning, alleviate pain, or prevent physical dysfunction in patients.
14CI 3–25 · exposure 13 · augmentation 50 · importance 4.8/5 · click for rater detail
Plan, prepare, or carry out individually designed programs of physical treatment to maintain, improve, or restore physical functioning, alleviate pain, or prevent physical dysfunction in patients.
14| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Healthcare generally lags in automation adoption; physical therapy clinics are small, diffuse, and slow to integrate new tools. Most adoption has been limited to back-office functions (scheduling, billing) rather than clinical workflows. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Healthcare delivery, especially hands-on rehabilitative care, is a slower-adopting sector for AI automation due to regulatory, safety, and physical-presence constraints. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can assist with program design via evidence synthesis, outcome tracking, and progress visualization to inform clinician decisions, moderately raising therapist productivity in the planning phase, though the hands-on treatment and adaptation remain clinician-driven. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI can assist with generating exercise plans, tracking progress, and analyzing movement data via sensors, but the core hands-on treatment execution remains untouched by AI assistance. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can assist with program planning through evidence-based guidelines and outcome tracking, the core task requires manual hands-on treatment, real-time assessment of patient response, and adaptive modification based on individual body mechanics and pain response—all fundamentally dependent on human clinician judgment and physical interaction that current AI cannot perform end-to-end. |
| Task automatability | claude-sonnet-5 | 1/5 | This task requires hands-on physical manipulation, real-time assessment of patient response, and manual therapy techniques that current AI systems cannot physically perform. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Physical therapy is a licensed profession with regulatory requirements that the treating clinician must hold a credential and legal responsibility for patient outcomes; liability and safety concerns create strong barriers to full automation, and most jurisdictions require direct human assessment and treatment. |
| Adoption barriers | claude-sonnet-5 | 5/5 | Physical therapy requires state licensure, direct physical contact, and legal accountability for patient safety, making autonomous AI substitution essentially prohibited. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | Current AI applications (scheduling, documentation, outcome tracking) are far cheaper than human therapists but address only marginal, preparatory parts of the task; the hands-on clinical work remains human-dependent, so any cost advantage is minimal relative to total service cost. |
| Cost vs. human wage | claude-sonnet-5 | 1/5 | Since AI cannot perform the physical, hands-on components at all, there is no viable cost comparison for full task substitution—human labor remains necessary. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | No deployed AI system can reliably perform the full spectrum of physical treatment (manipulation, exercise instruction, pain assessment, real-time modification). AI tools exist for documentation and outcome measurement, but these are narrow adjuncts, not solutions for the central clinical work. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product can carry out hands-on physical treatment; AI is not embodied in a way that allows physical manipulation of patients in clinical practice. |
Provide information to the patient about the proposed intervention, its material risks and expected benefits, and any reasonable alternatives.
13CI 0–25 · exposure 13 · augmentation 50 · importance 4.4/5 · click for rater detail
Provide information to the patient about the proposed intervention, its material risks and expected benefits, and any reasonable alternatives.
13| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 1/5 | Healthcare sectors, especially physical therapy, remain conservative on automating patient-facing informed-consent processes; clinical liability and patient-relationship expectations strongly inhibit adoption of AI-only disclosure. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Healthcare, especially hands-on rehabilitative care, has historically been slower to adopt AI for patient-facing clinical communication compared to purely administrative or information-sector tasks. |
| Augmentation potential | claude-haiku-4-5-20251001 | 2/5 | AI could draft consent documentation or outline standard risks and alternatives to assist the therapist's preparation, but the core task—direct patient communication, addressing individual concerns, and obtaining genuine informed consent—remains fundamentally human and cannot be substantially augmented by current AI. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI tools can meaningfully assist therapists by generating tailored patient education materials, summarizing risks/benefits, and supporting documentation, improving efficiency while the therapist remains the communicator of record. |
| Task automatability | claude-haiku-4-5-20251001 | 1/5 | This task requires real-time patient interaction, nuanced risk-benefit communication tailored to individual medical history, and establishment of informed consent—a fundamentally human-centered clinical judgment that current AI cannot perform end-to-end with 50% time savings at equal quality. |
| Task automatability | claude-sonnet-5 | 2/5 | AI can help draft patient-facing explanations of interventions, risks, and alternatives, but the actual patient conversation requires live clinical judgment, personalized risk communication, and legal/informed-consent responsibility that current systems cannot fully replace end-to-end. |
| Adoption barriers | claude-haiku-4-5-20251001 | 5/5 | Informed consent is a legally mandated duty of the licensed physical therapist; regulatory frameworks (state licensure, ADA, HIPAA) require the human provider to personally deliver material risk information and document patient understanding before treatment. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Informed consent and patient communication about risks/benefits is a core clinical and legal responsibility typically requiring the licensed provider's direct involvement, creating strong regulatory and liability barriers to full automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 1/5 | Any AI-assisted or automated attempt would require clinical supervision, patient verification, and liability oversight that negates cost advantage; human therapists remain cheaper when full compliance and malpractice risk are factored in. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | While generating educational content is cheap, the human clinician still must deliver, verify, and take liability for the information, so overall cost savings versus a licensed therapist doing this personally are modest. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 1/5 | No deployed product reliably performs informed-consent delivery for physical therapy interventions in production; AI systems lack the legal standing, liability authority, and patient-specific clinical reasoning integration required to discharge this duty of care. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Some patient-education chatbots and documentation tools exist, but no deployed product reliably conducts informed-consent-style risk/benefit/alternatives conversations in clinical PT practice at scale. |
Evaluate effects of treatment at various stages and adjust treatments to achieve maximum benefit.
10CI 0–20 · exposure 13 · augmentation 50 · importance 4.7/5 · click for rater detail
Evaluate effects of treatment at various stages and adjust treatments to achieve maximum benefit.
10| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 1/5 | Healthcare, especially physical therapy, remains in the laggard tier for AI automation due to regulatory constraints, patient safety concerns, and the hands-on nature of the work. Adoption is concentrated in measurement aids and documentation assistance rather than autonomous clinical decision-making. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Healthcare, especially hands-on physical therapy, has lower AI adoption due to regulatory constraints, physical interaction requirements, and slower digitization compared to information-based sectors. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can meaningfully assist by automating patient data collection, analyzing movement via video, tracking progress metrics, and flagging patterns that warrant clinical review. However, the therapist must still interpret context and make the final adjustment decision, so augmentation is moderate rather than transformative. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI can help by analyzing patient data trends, tracking outcome metrics, and suggesting evidence-based treatment modifications, supporting the therapist's decision-making process. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | Physical therapy evaluation requires ongoing clinical judgment, patient-specific assessment, and real-time observation of movement patterns and pain response. While AI can assist with measurement and data tracking, the core task of integrating subjective patient feedback with clinical reasoning to adjust treatments remains heavily dependent on human expertise and physical presence. |
| Task automatability | claude-sonnet-5 | 1/5 | This requires hands-on physical assessment, real-time observation of patient movement, pain response, and clinical judgment to modify treatment plans, which current AI cannot perform end-to-end. |
| Adoption barriers | claude-haiku-4-5-20251001 | 5/5 | Physical therapists are licensed professionals, and any autonomous adjustment of patient treatment plans would face strong legal and regulatory barriers. Patient safety liability, the requirement for licensed sign-off, and state-level scope-of-practice restrictions create hard barriers to substitution. |
| Adoption barriers | claude-sonnet-5 | 5/5 | Physical therapy treatment modification requires a licensed PT's professional judgment and is subject to strict licensing, liability, and regulatory requirements for patient care decisions. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI systems for physical therapy evaluation require significant integration (biomechanical sensors, video analysis pipelines, data management) and ongoing clinical oversight. The all-in cost per adjusted treatment plan remains substantially higher than or comparable to the labor cost of a licensed therapist performing the evaluation. |
| Cost vs. human wage | claude-sonnet-5 | 1/5 | AI cannot substitute for the physical examination and clinical decision-making involved, so there is no meaningful AI cost comparison for full task completion. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | Current AI tools can process patient data and suggest general protocols, but no deployed system reliably performs the full evaluation-and-adjustment cycle at production scale in clinical settings. AI excels at measurement but lacks the tactile feedback, movement analysis, and real-time clinical decision-making that define this task. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product independently evaluates physical therapy treatment progress and adjusts hands-on interventions; this remains a clinical judgment task performed by licensed professionals. |
Evaluate, fit, or adjust prosthetic or orthotic devices or recommend modification to orthotist.
9CI 3–16 · exposure 8 · augmentation 50 · importance 3.8/5 · click for rater detail
Evaluate, fit, or adjust prosthetic or orthotic devices or recommend modification to orthotist.
9| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Physical therapy is a regulated, human-contact-intensive field with slow digital adoption compared to information-based professions. While some clinics use measurement software, replacement of the core fitting and adjustment function is not occurring in production settings. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Healthcare and physical rehabilitation sectors show slow AI adoption for hands-on clinical tasks, though administrative and documentation aspects see more uptake. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can meaningfully assist by providing gait analysis, device fit visualization, or predictive recommendations for modifications, helping therapists make better decisions faster. However, the human therapist remains essential for the final evaluation and hands-on adjustment. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI can assist with documentation, tracking outcomes, analyzing gait data from sensors, or suggesting modifications based on data patterns, but the core physical fitting task still requires human execution. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | AI cannot perform the core task of physically evaluating, fitting, or adjusting prosthetic/orthotic devices, which requires hands-on manipulation, real-time patient feedback, and precise mechanical adjustment. AI could potentially assist in recommending modifications via image analysis or gait assessment, but only a narrow subset that falls well short of 50% time savings on the full task. |
| Task automatability | claude-sonnet-5 | 1/5 | This requires hands-on physical examination, tactile assessment of fit, gait analysis, and physical manipulation of devices on a patient's body—none of which current AI can perform end-to-end. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Significant regulatory barriers exist: physical therapists are licensed professionals, and patient safety liability for device fitting is high. Direct patient contact and licensed professional oversight are typically required, creating strong legal and regulatory protections against full automation. |
| Adoption barriers | claude-sonnet-5 | 5/5 | This requires a licensed physical therapist's clinical judgment and hands-on assessment, with liability and patient safety concerns requiring direct human evaluation and physical contact. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 1/5 | The physical manipulation and specialized expertise required mean AI cannot yet provide meaningful cost savings; the therapist's expertise and hands-on presence remain essential and remain far cheaper than developing and maintaining specialized robotic or AI systems for this work. |
| Cost vs. human wage | claude-sonnet-5 | 1/5 | AI cannot substitute for the physical labor involved, so there is no meaningful AI cost comparison—human clinicians remain necessary for the physical components. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 1/5 | No deployed AI product reliably performs physical evaluation, fitting, or adjustment of prosthetics/orthotics. Emerging computer vision for gait analysis exists in research, but production systems do not independently handle the mechanical fitting and real-time adjustment this task demands. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed AI product performs physical fitting or adjustment of prosthetic/orthotic devices; this remains a manual clinical task requiring physical presence. |
Inform patients and refer to appropriate practitioners when diagnosis reveals findings outside physical therapy.
6CI 0–11 · exposure 5 · augmentation 50 · importance 4.4/5 · click for rater detail
Inform patients and refer to appropriate practitioners when diagnosis reveals findings outside physical therapy.
6| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 1/5 | Healthcare remains a laggard in AI automation due to regulatory, liability, and professional licensing constraints; clinical triage tasks in particular are rarely automated, and when AI is used it is in a narrow support role under clinician control. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Healthcare, especially direct clinical diagnostic and referral decisions, adopts AI cautiously due to regulatory, liability, and safety concerns despite AI's use in adjacent documentation tasks. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can assist by flagging potential red flags or suggesting relevant specialists based on findings, helping a PT work faster and more comprehensively, though the final clinical decision and referral remains the therapist's responsibility. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI can help with differential diagnosis reference information, documentation of referral rationale, and identifying relevant specialists, supporting but not replacing the clinician's judgment. |
| Task automatability | claude-haiku-4-5-20251001 | 1/5 | This task requires clinical judgment to interpret diagnostic findings, recognize when they fall outside PT scope, and make appropriate referrals—decisions deeply rooted in human medical expertise and patient context that current AI cannot reliably perform end-to-end. |
| Task automatability | claude-sonnet-5 | 1/5 | This requires hands-on clinical assessment, direct patient interaction, and licensed clinical judgment about referral decisions that AI cannot perform end-to-end today. |
| Adoption barriers | claude-haiku-4-5-20251001 | 5/5 | This task is legally and ethically protected: a licensed physical therapist must make the clinical judgment and referral decision; liability and scope-of-practice regulations require human professional accountability and cannot be delegated to AI systems. |
| Adoption barriers | claude-sonnet-5 | 5/5 | This is a licensed clinical judgment task with legal scope-of-practice requirements; PTs must identify red flags and refer appropriately, a duty that cannot be legally delegated to software. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI decision-support tools exist but require substantial clinical oversight and integration; the cost per task (inference, validation, liability management) approaches or exceeds the marginal cost of a PT performing the judgment directly. |
| Cost vs. human wage | claude-sonnet-5 | 1/5 | AI cannot substitute for the licensed clinical evaluation and referral decision, so there is no viable AI-only cost comparison; the human must remain the decision-maker. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | While AI can assist in identifying certain out-of-scope findings via text analysis or decision support, no deployed product reliably performs the full task of clinical triage and referral independently; human clinicians must retain accountability for these decisions. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product autonomously conducts physical therapy diagnosis, identifies out-of-scope findings, and executes patient referrals in clinical practice. |
Discharge patient from physical therapy when goals or projected outcomes have been attained and provide for appropriate follow-up care or referrals.
4CI 0–7 · exposure 5 · augmentation 50 · importance 4.3/5 · click for rater detail
Discharge patient from physical therapy when goals or projected outcomes have been attained and provide for appropriate follow-up care or referrals.
4| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 1/5 | Healthcare remains a slow-adopter sector for autonomous clinical decision-making. Discharge decisions are high-touch and regulated; adoption of AI for this specific task is minimal despite AI investment in healthcare more broadly. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Healthcare and rehabilitative therapy sectors are slower AI adopters overall, with clinical decision-making tasks seeing pilots but minimal production deployment. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can helpfully summarize patient progress metrics, highlight goals met, and suggest referrals based on diagnoses, assisting the PT in structuring the discharge decision. However, the human PT retains full responsibility for the clinical judgment and final authorization. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI can help track outcome measures, summarize progress data, and draft referral letters or discharge summaries, meaningfully assisting documentation and follow-up planning around the human's judgment. |
| Task automatability | claude-haiku-4-5-20251001 | 1/5 | Discharging a patient requires clinical judgment about goal attainment, assessment of functional progress, and personalized follow-up decisions that depend on complex patient context. Current AI cannot reliably make these autonomous clinical discharge decisions or determine appropriate referrals without human oversight. |
| Task automatability | claude-sonnet-5 | 1/5 | Discharge decisions require clinical judgment about goal attainment, functional status, and safe transition planning that AI cannot perform end-to-end today.' |
| Adoption barriers | claude-haiku-4-5-20251001 | 5/5 | Physical therapists are licensed professionals; discharge decisions involve clinical judgment that regulatory bodies and liability frameworks require a credentialed human to document and authorize. Legal and professional standards create hard barriers to automation. |
| Adoption barriers | claude-sonnet-5 | 5/5 | Discharge and referral decisions are part of licensed clinical practice, requiring the PT's professional judgment and sign-off, with liability and regulatory frameworks tying this authority to a licensed provider. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 1/5 | The cost of an AI system for autonomous clinical discharge (including liability, integration, regulatory compliance, and oversight) far exceeds the cost of a PT spending 10–15 minutes to discharge a patient, especially given error and malpractice risk. |
| Cost vs. human wage | claude-sonnet-5 | 1/5 | Since AI cannot substitute for the clinical decision itself, there is no viable AI-only cost comparison; any AI use is supplementary to the human, not a replacement of the billed service. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | While AI can assist with documentation and flag patients meeting objective outcome metrics, no deployed product reliably performs autonomous patient discharge decisions in clinical settings. Regulatory and liability concerns mean this remains a human-driven task with limited AI autonomy in production. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product independently makes discharge determinations or coordinates follow-up referrals for physical therapy patients; this remains a clinician task. |
Obtain patients' informed consent to proposed interventions.
3CI 0–6 · exposure 0 · augmentation 38 · importance 4.7/5 · click for rater detail
Obtain patients' informed consent to proposed interventions.
3| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 1/5 | This task is bound by legal and ethical requirements that have not changed; physical therapy practices are not adopting AI for consent-taking because it is not legally permissible. Adoption velocity reflects regulatory immobility, not technology readiness. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Healthcare adoption of AI for clinical documentation is growing, but direct patient consent interactions remain untouched by automation due to legal and trust requirements. |
| Augmentation potential | claude-haiku-4-5-20251001 | 2/5 | AI could assist by pre-drafting consent forms or summarizing information, but the actual consent conversation and verification of patient understanding must remain with the licensed therapist. Assistance is marginal because the therapist cannot outsource the core accountability. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI can help generate plain-language consent materials, translate documents, or pre-fill forms, moderately assisting therapists in preparing for the consent conversation. |
| Task automatability | claude-haiku-4-5-20251001 | 1/5 | Informed consent requires a licensed professional to explain risks, benefits, and alternatives in a comprehensible way and assess patient understanding and voluntary agreement. This involves legal and ethical accountability that cannot be delegated to AI; current systems cannot reliably execute the legal gatekeeping function or handle the nuanced clinical judgment required. |
| Task automatability | claude-sonnet-5 | 1/5 | Obtaining informed consent requires a licensed clinician to explain risks, answer patient-specific questions, and confirm genuine understanding and voluntary agreement, which AI cannot legally or practically perform end-to-end today. |
| Adoption barriers | claude-haiku-4-5-20251001 | 5/5 | Informed consent is a legal and fiduciary obligation; a licensed physical therapist must personally obtain it and document the patient's understanding. Regulatory bodies, state licensing boards, and malpractice law all require a human professional to perform this gatekeeping function. Substitution is prohibited. |
| Adoption barriers | claude-sonnet-5 | 5/5 | Informed consent is a legally mandated clinical and ethical responsibility requiring a licensed provider's direct interaction and signature, making this one of the most protected tasks. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 1/5 | The cost of an AI system managing liability, errors, and oversight for consent (including malpractice exposure and regulatory compliance) would exceed the time cost of a physical therapist obtaining consent directly, since the task is already brief and non-automatable without removing human accountability. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | While drafting consent language is cheap via AI, the actual consent-obtaining interaction still requires a paid clinician's time, so overall cost savings are minimal. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 1/5 | No deployed AI system performs genuine informed consent in clinical practice. While chatbots can recite information, they cannot legally substitute for a licensed therapist obtaining consent or handle the interactive assessment of patient comprehension and autonomy that the law and ethics demand. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed products handle informed consent conversations autonomously for physical therapy interventions; at best AI produces consent-form text or educational materials for clinician use. |
Participate in community or community agency activities or help to formulate public policy.
3CI 0–5 · exposure 0 · augmentation 38 · importance 3.2/5 · click for rater detail
Participate in community or community agency activities or help to formulate public policy.
3| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 1/5 | This task involves professional representation and policy work that sectors actively protect as human responsibilities. There is no measurable adoption of AI agents in public policy or community agency participation roles. |
| Sector adoption velocity | claude-sonnet-5 | 1/5 | Civic and policy participation activities in healthcare professions show minimal AI adoption; this is a low-digitization, relationship-based activity. |
| Augmentation potential | claude-haiku-4-5-20251001 | 2/5 | AI could assist by drafting background research summaries or policy briefs, but the core task of participation and relationship-building cannot be meaningfully augmented by AI; the human must remain front-and-center. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI can help draft policy position papers, summarize community health data, or prepare talking points, providing moderate assistance to the human who still leads engagement. |
| Task automatability | claude-haiku-4-5-20251001 | 1/5 | This task requires sustained engagement with community stakeholders, policy deliberation, and advocacy grounded in professional expertise and contextual judgment. Current AI systems cannot autonomously participate in community meetings, represent professional interests, or author defensible policy recommendations without human leadership. |
| Task automatability | claude-sonnet-5 | 1/5 | This task requires in-person community engagement, relationship building, and human judgment about local needs and policy advocacy that AI cannot perform end-to-end."}, |
| Adoption barriers | claude-haiku-4-5-20251001 | 5/5 | Community participation and policy formulation are inherently human civic and professional responsibilities; there are legal, ethical, and regulatory expectations that a licensed physical therapist—not an AI—represents the profession and constituency in these forums. Liability and accountability rest with the human. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Community representation and policy advocacy typically require a credentialed, accountable human representative with professional standing and trust, creating strong social and institutional barriers to automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 1/5 | Physical therapists earn substantial salaries (>$70k/yr loaded), and AI systems offer no cost advantage for this task since they cannot perform it at all. Human involvement is mandatory. |
| Cost vs. human wage | claude-sonnet-5 | 1/5 | There is no AI substitute performing this task, so cost comparison favors the human by default since AI cannot deliver the output at all. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 1/5 | No deployed product performs community participation or policy formulation autonomously. AI tools may draft talking points or summarize policy documents, but actual participation in agency activities and policy development remains entirely human-driven in practice. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed AI product participates in community organizations or formulates public policy on behalf of a professional; this remains entirely human-driven civic activity. |
Confer with the patient, medical practitioners, or appropriate others to plan, implement, or assess the intervention program.
1CI 0–3 · exposure 0 · augmentation 50 · importance 4.7/5 · click for rater detail
Confer with the patient, medical practitioners, or appropriate others to plan, implement, or assess the intervention program.
1| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 1/5 | Healthcare remains a heavily regulated, human-contact-intensive sector with slow digital AI adoption for clinical decision roles. Conferencing and care planning are not areas where AI substitution has visibly gained traction in production settings. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Healthcare, especially hands-on clinical fields like physical therapy, adopts AI slowly for core clinical decision-making tasks due to regulation and interpersonal requirements, though documentation tools are spreading. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can assist by summarizing patient history, suggesting evidence-based intervention options, or drafting meeting notes, reducing clerical burden. However, the core collaborative and clinical judgment elements remain human-driven, limiting the transformative potential of assistance. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI can assist by summarizing patient records, drafting notes, or suggesting evidence-based intervention options, aiding the therapist's conferral and planning process without replacing the interaction. |
| Task automatability | claude-haiku-4-5-20251001 | 1/5 | This task requires real-time interpersonal negotiation, clinical judgment synthesis across multiple stakeholders, and adaptive communication that current AI cannot reliably perform. The coordination of conflicting medical perspectives and patient-specific contextual concerns falls far outside what AI can do end-to-end today. |
| Task automatability | claude-sonnet-5 | 1/5 | This requires real-time interpersonal conferral, clinical judgment, and relationship-building with patients and other clinicians that current AI cannot substitute end-to-end. |
| Adoption barriers | claude-haiku-4-5-20251001 | 5/5 | Physical therapists are licensed professionals; patient care conferencing and treatment planning are legally scope-of-practice activities that require a licensed human to participate, document, and sign off. Liability and regulatory requirements create hard barriers to full automation. |
| Adoption barriers | claude-sonnet-5 | 5/5 | Physical therapy licensure, scope-of-practice laws, and liability requirements mandate that a qualified human professional confer with patients and other practitioners on treatment plans. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 1/5 | The overhead of AI systems to generate meaningful clinical input, combined with necessary human oversight and verification of any output, makes the all-in cost exceed what a physical therapist would charge for direct conferencing. |
| Cost vs. human wage | claude-sonnet-5 | 1/5 | Since AI cannot perform this collaborative clinical task independently, there is no viable AI-only cost comparison; a licensed therapist's involvement remains necessary. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 1/5 | No deployed product reliably performs the full conferencing, planning, and assessment cycle with medical practitioners and patients in real clinical settings. While AI can draft notes or summarize records, the actual collaborative decision-making and relationship-building remain absent from production systems. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product autonomously conducts multi-party clinical care conferences or treatment planning discussions in production settings today. |
Direct, supervise, assess, and communicate with supportive personnel.
1CI 0–3 · exposure 0 · augmentation 38 · importance 4.6/5 · click for rater detail
Direct, supervise, assess, and communicate with supportive personnel.
1| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 1/5 | Healthcare sectors have strict regulatory constraints on delegation of clinical supervision duties. There is no meaningful adoption of AI systems replacing therapist supervision in production settings. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Healthcare adoption of AI for clinical/managerial supervision tasks is slow due to regulatory and liability constraints, though documentation-adjacent tools are spreading. |
| Augmentation potential | claude-haiku-4-5-20251001 | 2/5 | AI could offer limited assistance with documentation of assessments or scheduling of supervised sessions, but the core supervisory and assessment functions remain dependent on the therapist's clinical judgment and presence. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI can help track staff performance metrics, scheduling, and documentation to support supervision, but the core communication and assessment remains human-led. |
| Task automatability | claude-haiku-4-5-20251001 | 1/5 | This task fundamentally requires human judgment to assess patient condition, direct clinical staff decisions, and make real-time therapeutic adjustments. Current AI systems cannot reliably supervise or make clinical decisions that depend on nuanced observation and responsiveness to individual patient needs. |
| Task automatability | claude-sonnet-5 | 1/5 | Supervising and assessing human staff requires interpersonal judgment, accountability, and real-time management that AI cannot perform end-to-end today. |
| Adoption barriers | claude-haiku-4-5-20251001 | 5/5 | Physical therapists must be licensed healthcare professionals who legally sign off on patient treatment plans and supervise personnel. This creates a hard regulatory barrier that prevents automation of clinical supervision and assessment. |
| Adoption barriers | claude-sonnet-5 | 5/5 | Licensed PTs are legally and professionally responsible for supervising supportive personnel (e.g., PTAs, aides), a non-delegable regulatory and liability requirement. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 1/5 | AI offers no meaningful cost advantage here because human supervision with clinical accountability remains legally and ethically mandatory. Any AI assistance would add cost rather than replace the supervised professional. |
| Cost vs. human wage | claude-sonnet-5 | 1/5 | There is no viable AI substitute performing this supervisory task, so cost comparison favors the human by default. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 1/5 | No deployed AI product can perform clinical supervision, patient assessment, or personnel management in physical therapy settings. These require licensed professionals with accountability for patient safety and staff performance. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed AI product supervises or directs clinical support personnel; this remains a human management function. |
Administer manual exercises, massage, or traction to help relieve pain, increase patient strength, or decrease or prevent deformity or crippling.
0CI 0–0 · exposure 0 · augmentation 25 · importance 4.7/5 · click for rater detail
Administer manual exercises, massage, or traction to help relieve pain, increase patient strength, or decrease or prevent deformity or crippling.
0| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 1/5 | Physical therapy remains a fundamentally hands-on, human-contact-required service in low-digitization sectors (small clinics, rehabilitation centers). There is minimal evidence of AI agent adoption displacing this task in production; the sector lags in automation overall. |
| Sector adoption velocity | claude-sonnet-5 | 1/5 | Healthcare, especially hands-on physical therapy, is a slow-adopting, highly regulated, physically-mediated sector with minimal AI displacement of manual treatment tasks. |
| Augmentation potential | claude-haiku-4-5-20251001 | 2/5 | AI can support physical therapists by optimizing exercise plans, analyzing patient progress data, or recommending adjustments based on research literature. However, the core manual and tactile delivery of therapy cannot be augmented by current AI in ways that materially raise therapist productivity during patient treatment itself. |
| Augmentation potential | claude-sonnet-5 | 2/5 | AI can assist with exercise planning, tracking progress, or generating home exercise programs, but offers little direct augmentation of the hands-on manual technique itself. |
| Task automatability | claude-haiku-4-5-20251001 | 1/5 | This task requires direct physical manipulation of a patient's body with real-time tactile feedback and adjustment based on patient response. Current AI systems cannot physically perform hands-on manual therapy, massage, or traction, and no robotic system in production reliably substitutes for a trained physical therapist's touch and judgment. |
| Task automatability | claude-sonnet-5 | 1/5 | This requires direct physical manipulation of a patient's body—manual exercises, massage, and traction—which current AI systems cannot physically perform, lacking embodiment and dexterity. |
| Adoption barriers | claude-haiku-4-5-20251001 | 5/5 | Physical therapy is a licensed profession; only licensed physical therapists or their direct supervision can legally administer these treatments in most jurisdictions. Additionally, the requirement for hands-on patient contact and real-time clinical judgment creates a hard regulatory barrier to full automation. |
| Adoption barriers | claude-sonnet-5 | 5/5 | Physical therapy is a licensed profession with legal requirements for a qualified human to administer hands-on treatment, plus liability concerns around physical harm. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 1/5 | Specialized robotic systems capable of any aspect of manual therapy are capital-intensive, require technical support and maintenance, and still demand human supervision. The all-in cost far exceeds the hourly rate of a physical therapist for actual therapeutic delivery. |
| Cost vs. human wage | claude-sonnet-5 | 1/5 | There is no viable AI substitute for hands-on manual therapy, so any hypothetical robotic system would be far more expensive than a human therapist's labor for this task. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 1/5 | While robotic arms exist in research and some specialized medical settings, no deployed product reliably performs manual exercise prescription, massage, or traction administration at the quality and safety standards required for independent patient care. Deployed systems remain research-stage or require extensive human oversight. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product performs hands-on manual therapy; robotic physical therapy aids exist only in narrow research/rehab contexts, not as substitutes for manual treatment. |
Administer treatment involving application of physical agents, using equipment, moist packs, ultraviolet or infrared lamps, or ultrasound machines.
0CI 0–0 · exposure 0 · augmentation 25 · importance 3.9/5 · click for rater detail
Administer treatment involving application of physical agents, using equipment, moist packs, ultraviolet or infrared lamps, or ultrasound machines.
0| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 1/5 | Physical therapy is a hands-on, client-facing profession with minimal digital-first adoption; the sector remains heavily dependent on in-person clinician-patient interaction with little evidence of displacement by automation. |
| Sector adoption velocity | claude-sonnet-5 | 1/5 | Healthcare physical treatment delivery is a low-digitization, hands-on sector with minimal AI/robotic adoption for direct patient-contact modality application. |
| Augmentation potential | claude-haiku-4-5-20251001 | 2/5 | AI could assist with treatment planning or patient education, but the core manual application of physical agents offers limited opportunity for meaningful AI augmentation while the therapist remains in the loop. |
| Augmentation potential | claude-sonnet-5 | 2/5 | AI can help with treatment planning, documentation, or monitoring equipment settings, but offers little direct assistance during the physical application of the treatment itself. |
| Task automatability | claude-haiku-4-5-20251001 | 1/5 | This task requires hands-on physical manipulation of equipment and direct application of therapeutic agents to patients' bodies, necessitating real-time tactile feedback and precise positioning that current AI systems cannot perform autonomously in a clinical setting. |
| Task automatability | claude-sonnet-5 | 1/5 | This requires hands-on physical manipulation of equipment on a patient's body, positioning, and real-time adjustment based on tactile/visual feedback that current AI systems cannot perform without robotic embodiment. |
| Adoption barriers | claude-haiku-4-5-20251001 | 5/5 | Physical therapy treatment administration is a licensed clinical activity requiring a licensed physical therapist; medical liability, patient safety regulations, and direct human-contact requirements create strong legal and regulatory barriers to automation. |
| Adoption barriers | claude-sonnet-5 | 5/5 | Physical therapy treatment application is a licensed clinical activity requiring hands-on delivery, direct supervision, and legal accountability, creating strong regulatory and liability barriers to automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 1/5 | The cost of robotic systems capable of safely manipulating therapeutic equipment on patients, plus integration and safety oversight, far exceeds the loaded wage of a physical therapist performing this task. |
| Cost vs. human wage | claude-sonnet-5 | 1/5 | There is no viable AI substitute performing this physical task, so any AI cost comparison is moot and the human remains the only option. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 1/5 | No deployed AI product can independently administer physical therapy treatments involving direct patient contact, equipment operation, and application of therapeutic modalities; this remains a human-performed clinical task. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product administers physical modality treatments like ultrasound or hot/cold packs autonomously in clinical practice; this remains purely manual clinical work. |
Direct group rehabilitation activities.
0CI 0–0 · exposure 0 · augmentation 25 · importance 3.1/5 · click for rater detail
Direct group rehabilitation activities.
0| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 1/5 | Healthcare adoption of AI for core clinical tasks remains slow and heavily regulated; physical therapy is a hands-on, relationship-intensive field where human presence is embedded in reimbursement, licensing, and care standards, limiting any velocity toward automation. |
| Sector adoption velocity | claude-sonnet-5 | 1/5 | Healthcare/rehabilitation is a physically-grounded, highly regulated sector with slow AI adoption for hands-on clinical delivery tasks. |
| Augmentation potential | claude-haiku-4-5-20251001 | 2/5 | AI could modestly assist by recording exercise form for offline review or tracking aggregate group progress metrics, but such tools offer limited real-time value during live group direction and do not meaningfully amplify the therapist's core task of dynamic supervision and patient engagement. |
| Augmentation potential | claude-sonnet-5 | 2/5 | AI can help plan exercise protocols or track progress data, but offers limited real-time assistance during the actual directing of group sessions. |
| Task automatability | claude-haiku-4-5-20251001 | 1/5 | Directing group rehabilitation activities requires real-time physical assessment, dynamic adaptation to individual patient capabilities, motivational coaching, and hands-on correction of movement—all deeply dependent on human presence, observation, and interpersonal judgment that current AI systems cannot perform end-to-end. |
| Task automatability | claude-sonnet-5 | 1/5 | Directing group rehabilitation requires live physical observation, hands-on correction, and real-time adaptation to multiple patients' movements and safety needs, which current AI cannot perform end-to-end. |
| Adoption barriers | claude-haiku-4-5-20251001 | 5/5 | Physical therapists must be state-licensed and legally responsible for patient safety, assessment, and treatment outcomes; regulations explicitly require qualified human professionals to direct rehabilitation and make clinical decisions, creating a hard legal and liability barrier to automation. |
| Adoption barriers | claude-sonnet-5 | 5/5 | Physical therapy is a licensed profession with legal requirements for a qualified therapist to supervise treatment, and liability for patient injury during exercise is high. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 1/5 | The cost of any AI system capable of monitoring and directing group rehabilitation (computer vision, sensor arrays, real-time feedback systems) would far exceed the hourly cost of a licensed physical therapist, and the technology does not yet exist at clinical-grade reliability. |
| Cost vs. human wage | claude-sonnet-5 | 1/5 | There is no viable AI substitute performing this task, so cost comparison favors the human clinician entirely. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 1/5 | No deployed AI product reliably directs group rehabilitation activities today; this task demands live supervision, physical intervention, and individualized real-time feedback that existing systems cannot replicate at the necessary reliability for a clinical setting. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product independently directs group physical rehabilitation sessions in clinical settings; this remains outside current AI product 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.