Hospitalists

29-1229.02
Median wage $265,930/yr342,720 employed (US)Rank #772 of 923 scored · top 84% by substitution

Provide inpatient care predominantly in settings such as medical wards, acute care units, intensive care units, rehabilitation centers, or emergency rooms. Manage and coordinate patient care throughout treatment.

Sub-scores

0–100 · band = confidence interval from rater disagreement

Substitution16
Exposure16
Augmentation64

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

14 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%19

panel mean rating 1.8/5 → substitution pressure 19/100

Cost vs. human wagew 15%18

panel mean rating 1.7/5 → substitution pressure 18/100

Adoption barriersw 20%inverted — strong barriers lower the score9

panel mean rating 4.7/5 (barrier strength) → substitution pressure 9/100

Sector adoption velocityw 10%26

panel mean rating 2.1/5 → substitution pressure 26/100

Task breakdown (14 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.

Write patient discharge summaries and send them to primary care physicians.

53

CI 49–57 · exposure 58 · augmentation 88 · importance 4.7/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510013/5Hospitals are piloting clinical documentation AI, but deployment remains inconsistent; uptake is faster in larger health systems and academic centers than in smaller institutions, and production-scale integration with EHRs and inter-provider communication workflows is still maturing.
Sector adoption velocityclaude-sonnet-53/5Healthcare systems are piloting and increasingly deploying AI scribing/summarization tools, but overall clinical documentation AI adoption remains moderate compared to faster-moving sectors like finance or tech.
Augmentation potentialclaude-haiku-4-5-202510014/5AI significantly accelerates discharge summary drafting by auto-populating structured sections and suggesting clinical language, enabling hospitalists to review and refine rather than compose from scratch, materially raising throughput while preserving physician oversight and liability.
Augmentation potentialclaude-sonnet-55/5AI-generated draft summaries substantially reduce physician documentation time and cognitive load while the physician remains responsible for review and transmission, making this a strong augmentation use case.
Task automatabilityclaude-haiku-4-5-202510014/5AI language models can generate coherent discharge summaries from clinical notes with substantial time savings, though oversight and clinical validation remain necessary. Current systems reliably capture medication lists, diagnoses, and care instructions, reducing drafting time by 60–80%, meeting the ≥50% threshold with some human review.
Task automatabilityclaude-sonnet-53/5AI can draft discharge summaries from clinical notes with significant time savings, but a physician must review, verify accuracy, and finalize the clinical content before sending, limiting full end-to-end automation.'
Adoption barriersclaude-haiku-4-5-202510014/5Discharge summaries must be reviewed and signed by a physician (legal and malpractice liability), and many institutions maintain strict documentation standards and require direct clinical verification before transmission to primary care; these requirements prevent full automation without human sign-off.
Adoption barriersclaude-sonnet-54/5Discharge summaries are legal medical records requiring physician authorship/attestation and are subject to clinical liability and regulatory documentation standards, so a licensed physician must sign off.
Cost vs. human wageclaude-haiku-4-5-202510014/5AI inference and integration costs are typically a few dollars per summary, compared to 20–40 minutes of hospitalist time (~$15–30 per summary in loaded wages); the per-task cost is substantially lower even accounting for oversight.
Cost vs. human wageclaude-sonnet-54/5Generating a draft summary via LLM costs a fraction of a cent to a few dollars in compute versus substantial physician time, though oversight and correction costs reduce the net savings somewhat.
Technical feasibility todayclaude-haiku-4-5-202510013/5Several EHR-integrated and standalone products (e.g., clinical documentation AI, large-language-model-based note generators) exist and are deployed in hospitals, but error rates remain material (missed drug interactions, incomplete plan details) and clinical adoption is still inconsistent across institutions.
Technical feasibility todayclaude-sonnet-53/5Several EHR-integrated ambient documentation and summarization tools (e.g., Epic's AI features, Nuance DAX) generate discharge summary drafts in production, but adoption is uneven and human editing is standard practice.

Order or interpret the results of tests such as laboratory tests and radiographs (x-rays).

28

CI 20–36 · exposure 30 · augmentation 75 · importance 4.9/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510012/5While radiology AI tools are deployed in some health systems, adoption remains limited and primarily in advisory or quality-control roles, not replacement. Hospitalists continue to order and interpret tests as a core responsibility; displacement is minimal despite available technology.
Sector adoption velocityclaude-sonnet-53/5Hospitals are adopting AI diagnostic aids (radiology, sepsis alerts) at a moderate pace, with pilots and FDA-cleared tools increasingly used but still supplementary rather than primary decision tools.
Augmentation potentialclaude-haiku-4-5-202510014/5AI excels at pre-screening radiographs for critical findings, summarizing lab abnormalities, and flagging results requiring attention, meaningfully accelerating the physician's ability to triage and review. Hospitalists using AI-assisted interpretation tools are demonstrably more efficient and less likely to miss findings.
Augmentation potentialclaude-sonnet-54/5AI significantly aids hospitalists by pre-flagging abnormal results, prioritizing critical findings, and providing decision support, meaningfully speeding up review and reducing missed findings while the physician retains final judgment.
Task automatabilityclaude-haiku-4-5-202510012/5AI can assist in flagging abnormal results and suggesting diagnoses, but radiograph interpretation requires nuanced clinical judgment, integration with patient history, and accountability that current systems cannot fully automate. The task involves both ordering (with clinical reasoning) and interpretation (with liability), neither of which meets the 50% time-saving bar without significant human oversight remaining.
Task automatabilityclaude-sonnet-52/5AI can draft orders and flag abnormal lab values or suggest radiograph findings, but final ordering decisions and clinical interpretation integrated into patient management still require physician judgment and liability, so full end-to-end automation with equal quality is not yet achievable.
Adoption barriersclaude-haiku-4-5-202510014/5Ordering and interpreting tests carry direct patient safety and liability implications; regulatory bodies (FDA, state medical boards) and malpractice frameworks require a licensed physician to order, interpret, or sign off on diagnostic results. Automation without licensed human accountability is not legally permissible.
Adoption barriersclaude-sonnet-55/5Ordering and interpreting tests for diagnosis and treatment is a core licensed physician responsibility with direct liability exposure, requiring physician sign-off under medical practice regulations.
Cost vs. human wageclaude-haiku-4-5-202510012/5AI inference is cheap, but the task requires integration into EHR workflows, image handling infrastructure, and physician oversight. The all-in cost per interpretation remains higher than a cost-saving compared to a radiologist or hospitalist, especially when liability and verification labor are included.
Cost vs. human wageclaude-sonnet-53/5AI-assisted image triage and lab interpretation tools are relatively cheap per use, but the need for physician oversight and integration into EHR/order systems keeps overall cost roughly comparable to current physician time saved.
Technical feasibility todayclaude-haiku-4-5-202510013/5Commercial AI systems exist for radiology interpretation (e.g., detection of pneumonia on chest X-rays, flagging critical findings) and lab result analysis, but they operate within narrow scope, have non-trivial false-positive/false-negative rates, and require radiologist or physician review. Deployed products assist but do not independently assume accountability.
Technical feasibility todayclaude-sonnet-52/5Deployed AI radiology-assist tools (e.g., for fracture or nodule detection) and lab flagging systems exist but are narrow, adjunctive, and not autonomous decision-makers in hospitalist workflows.

Refer patients to medical specialists, social services, or other professionals as appropriate.

23

CI 20–25 · exposure 25 · augmentation 75 · importance 4.6/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510012/5Referral support tools see limited production adoption despite decades of clinical decision support research. Hospitals remain cautious about automation in referral pathways due to liability concerns and physician resistance to algorithmic oversight of professional judgment.
Sector adoption velocityclaude-sonnet-52/5Healthcare overall adopts AI slowly for clinical decision-making due to regulatory, liability, and EHR integration hurdles, though administrative-adjacent uses (documentation) are growing faster.
Augmentation potentialclaude-haiku-4-5-202510014/5AI systems meaningfully assist hospitalists by flagging specialty-specific guidelines, identifying patients who meet evidence-based referral criteria, and suggesting relevant services—raising productivity while the physician retains final decision authority. Current clinical decision support demonstrates clear assistive value in this workflow.
Augmentation potentialclaude-sonnet-54/5AI can meaningfully assist hospitalists by summarizing patient history, suggesting relevant specialists or resources, and auto-drafting referral paperwork, improving efficiency while the physician retains final decision authority.
Task automatabilityclaude-haiku-4-5-202510012/5While AI can assist in suggesting specialist referrals based on clinical guidelines and patient data, the task requires nuanced clinical judgment about appropriateness, timing, and patient-specific factors that current systems cannot reliably handle end-to-end. The decision to refer demands integration of complex patient context that exceeds 50% time savings at equal quality.
Task automatabilityclaude-sonnet-52/5Drafting referral letters or matching symptoms to specialties can be partially automated, but the actual judgment of when and to whom to refer requires clinical reasoning, patient-specific context, and accountability that current AI cannot reliably assume end-to-end.
Adoption barriersclaude-haiku-4-5-202510014/5Strong legal and regulatory barriers exist: physicians are legally responsible for referral appropriateness, and liability for missed or inappropriate referrals rests with the attending clinician. Professional standards and licensing requirements mandate human physician judgment for these decisions.
Adoption barriersclaude-sonnet-55/5Referrals are a licensed medical act tied to diagnosis and care coordination, requiring a physician's professional judgment and legal accountability, making this a hard-barrier task.
Cost vs. human wageclaude-haiku-4-5-202510012/5AI systems for clinical decision support require expensive integration with EHRs, ongoing model maintenance, and mandatory human oversight. The total cost per referral decision remains comparable to or exceeds the marginal cost of a hospitalist making the decision.
Cost vs. human wageclaude-sonnet-52/5AI can cheaply draft referral documentation, but the physician's judgment and liability exposure mean the human cost is not meaningfully displaced, keeping overall cost comparable to or only slightly better than human-only workflows.
Technical feasibility todayclaude-haiku-4-5-202510012/5Decision-support tools exist to flag potential referrals, but no deployed product reliably performs independent referral decisions in production settings. Products remain advisory rather than autonomous, with hospitals still requiring physician review and sign-off on all referral recommendations.
Technical feasibility todayclaude-sonnet-52/5Some clinical decision-support tools suggest specialist referrals or flag social needs, but no deployed product autonomously makes and executes referral decisions in production without physician review.

Communicate with patients' primary care physicians upon admission, when treatment plans change, or at discharge to maintain continuity and quality of care.

21

CI 16–25 · exposure 17 · augmentation 63 · importance 4.4/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510012/5Healthcare adoption of AI for clinical communication remains cautious and pilot-stage; liability concerns and regulatory uncertainty slow deployment. While some EHR vendors are adding AI documentation features, actual displacement of physician communication tasks remains minimal in production settings.
Sector adoption velocityclaude-sonnet-52/5Healthcare, especially inpatient clinical workflows, has been a laggard sector for AI adoption in actual physician communication tasks despite growing EHR-based documentation assistance.
Augmentation potentialclaude-haiku-4-5-202510013/5AI can help draft clinical summaries and extract key information from admission records to assist the hospitalist in preparing communications. However, augmentation is partial because the physician must still review, verify, and ultimately compose or approve the message to ensure clinical accuracy and completeness.
Augmentation potentialclaude-sonnet-54/5AI can meaningfully assist by auto-drafting discharge summaries, extracting relevant treatment plan changes, and flagging key information to relay to PCPs, improving efficiency while the physician remains responsible for the communication.
Task automatabilityclaude-haiku-4-5-202510011/5Communication with primary care physicians requires understanding complex clinical context, patient history, and nuanced judgment about what information is relevant. Current AI cannot reliably select, synthesize, and convey clinical details appropriately without human oversight.
Task automatabilityclaude-sonnet-52/5Drafting communication summaries can be automated, but synthesizing clinical judgment, timing communication appropriately, and handling nuanced discussions with PCPs requires human clinical reasoning and accountability that current AI cannot fully replace end-to-end.
Adoption barriersclaude-haiku-4-5-202510014/5Regulatory and liability barriers are high: a licensed physician must take responsibility for clinical communications, and any miscommunication between inpatient and primary care teams can have direct patient safety consequences. Medical licensing law typically requires a physician to personally participate in such handoffs.
Adoption barriersclaude-sonnet-54/5Physician-to-physician communication about patient care is generally required to be performed or verified by a licensed physician due to liability, continuity-of-care regulations, and medical malpractice exposure.
Cost vs. human wageclaude-haiku-4-5-202510012/5AI-assisted documentation tools exist but still require hospitalist review and approval. The total cost (AI tool subscriptions plus significant human oversight and validation) approaches but does not clearly undercut the loaded wage of a hospitalist or administrative staff member.
Cost vs. human wageclaude-sonnet-52/5While AI drafting tools reduce documentation time cheaply, the physician still must review, verify, and often personally communicate, so overall cost savings versus a hospitalist's time are modest rather than order-of-magnitude.
Technical feasibility todayclaude-haiku-4-5-202510012/5While AI can draft clinical notes and summaries, no deployed system reliably handles the full task of communicating with external physicians with appropriate clinical judgment. Experimental systems exist but lack production-scale reliability in real healthcare settings.
Technical feasibility todayclaude-sonnet-52/5Some EHR-integrated tools can auto-generate discharge summaries or care transition notes, but actual physician-to-physician communication and clinical dialogue is not reliably handled by deployed AI products today.

Direct or support quality improvement projects or safety programs.

19

CI 13–25 · exposure 17 · augmentation 63 · importance 4.0/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510012/5Healthcare adoption of AI remains cautious and heavily regulated; while some health systems pilot AI-assisted analytics, autonomous direction of safety programs is not yet being deployed in practice at scale.
Sector adoption velocityclaude-sonnet-52/5Healthcare quality improvement functions are adopting AI slowly, mostly for data analytics support rather than leadership or program direction.
Augmentation potentialclaude-haiku-4-5-202510013/5AI can usefully assist hospitalists by automating data aggregation, identifying safety trends from medical records, and generating reports on quality metrics, enabling faster analysis to inform their QI leadership decisions.
Augmentation potentialclaude-sonnet-54/5AI can meaningfully assist by analyzing safety incident data, identifying trends, drafting reports, and suggesting interventions, boosting hospitalist productivity in these projects.
Task automatabilityclaude-haiku-4-5-202510011/5Directing or supporting quality improvement and safety programs requires strategic decision-making, stakeholder engagement, change management, and judgment about organizational priorities and human factors. Current AI cannot autonomously lead such complex, human-centered initiatives.
Task automatabilityclaude-sonnet-52/5Leading and championing quality improvement or safety programs requires organizational judgment, stakeholder buy-in, and clinical context that current AI cannot autonomously execute end-to-end.'
Adoption barriersclaude-haiku-4-5-202510014/5Healthcare quality and safety programs carry significant liability, regulatory oversight (CMS, accreditation bodies), and require institutional accountability; a licensed clinician must be responsible for patient safety outcomes and program integrity.
Adoption barriersclaude-sonnet-54/5Quality and safety leadership roles typically require clinical licensure, institutional accountability, and regulatory/accreditation frameworks (e.g., Joint Commission) that mandate physician oversight.
Cost vs. human wageclaude-haiku-4-5-202510011/5AI assistance for QI support remains expensive relative to the incremental productivity gain, and hospitalists' labor is already highly specialized; the loaded human cost is low relative to AI infrastructure needed for autonomous direction.
Cost vs. human wageclaude-sonnet-52/5Human physician leadership time is expensive, but the analytical/data components AI can offer only partially substitute, so overall cost savings versus a hospitalist's involvement are modest.
Technical feasibility todayclaude-haiku-4-5-202510012/5While AI can assist with data analysis and reporting on safety metrics, no deployed product can autonomously direct or support QI/safety programs; existing tools support humans rather than performing the core leadership function.
Technical feasibility todayclaude-sonnet-52/5AI tools exist for analyzing safety data or drafting QI reports, but no deployed product independently directs or leads hospital quality improvement initiatives in production.

Participate in continuing education activities to maintain or enhance knowledge and skills.

16

CI 5–28 · exposure 17 · augmentation 63 · importance 4.3/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510011/5This task is inherently personal and legally mandated for individual practitioners; there is no sector-wide adoption pattern of automation because the task cannot be delegated away by design.
Sector adoption velocityclaude-sonnet-53/5Healthcare and medical education have moderate digitization with growing use of AI-driven learning platforms and adaptive quizzes, but full-scale replacement of CME participation is not underway.
Augmentation potentialclaude-haiku-4-5-202510013/5AI can meaningfully assist by recommending relevant courses, summarizing clinical literature, or organizing learning materials to help a hospitalist prepare for or maximize their continuing education activities, though the core participation remains human-driven.
Augmentation potentialclaude-sonnet-54/5AI can significantly enhance this task by curating relevant literature, summarizing new research, generating personalized learning modules, and quizzing physicians, improving efficiency of knowledge acquisition while the physician remains the learner of record.
Task automatabilityclaude-haiku-4-5-202510011/5Continuing education participation requires human judgment about skill gaps, motivation to learn, and engagement with material—none of which AI can autonomously perform on behalf of a hospitalist. AI cannot enroll in, attend, or meaningfully participate in the educational activities themselves.
Task automatabilityclaude-sonnet-52/5AI can deliver curated content and quizzes but cannot itself 'participate' in the learning process, maintain licensure credits, or provide the accreditation and hands-on clinical exposure required; the core activity is a human obligation.wagon.
Adoption barriersclaude-haiku-4-5-202510015/5Medical licensing bodies (state boards, specialty certifications) legally require hospitalists to maintain active participation in continuing education as a condition of licensure and board certification. No automated system can satisfy this regulatory requirement.
Adoption barriersclaude-sonnet-54/5Board certification and CME requirements are regulated by medical licensing bodies and specialty boards, mandating that the physician personally complete and attest to educational activities—this is a hard institutional/legal barrier to substitution.
Cost vs. human wageclaude-haiku-4-5-202510011/5AI cannot substitute for human participation in continuing education; the task requires credential maintenance and personal professional development that a human must perform. Any AI support is supplementary, not replacement.
Cost vs. human wageclaude-sonnet-52/5AI tools can cheaply generate study materials or summaries, but the accredited CME infrastructure (courses, testing, certification) still requires human-run programs, keeping overall costs comparable rather than dramatically lower.
Technical feasibility todayclaude-haiku-4-5-202510012/5While AI can assist with finding educational resources or summarizing content, no deployed product can actually participate in continuing education on a hospitalist's behalf. Some tools help identify relevant learning materials, but active participation remains fundamentally a human task.
Technical feasibility todayclaude-sonnet-52/5AI-powered learning platforms and summarization tools exist and are used for CME content delivery, but no product autonomously fulfills a physician's continuing education requirement end-to-end.

Conduct discharge planning and discharge patients.

16

CI 11–20 · exposure 17 · augmentation 75 · importance 4.8/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510012/5Healthcare adoption of AI-driven discharge automation is slow; most hospitals use AI only for documentation assistance or decision support, not autonomous execution. Regulatory caution, malpractice liability concerns, and institutional resistance to removing physician judgment from discharge planning limit uptake.
Sector adoption velocityclaude-sonnet-52/5Healthcare overall adopts AI more cautiously than finance or tech, and discharge decisions specifically remain a slow-adoption, high-scrutiny area with mostly pilot-stage tools.
Augmentation potentialclaude-haiku-4-5-202510014/5AI significantly assists hospitalists in discharge planning through rapid summarization of complex medical histories, identification of social/functional barriers, flagging of high-risk patients, and generation of discharge documentation. These augmentations measurably improve workflow efficiency and safety while the physician retains final authority and judgment.
Augmentation potentialclaude-sonnet-54/5AI tools meaningfully speed up drafting discharge summaries, checking medication interactions, and identifying readmission risks, substantially aiding the hospitalist's workflow while they retain final judgment.
Task automatabilityclaude-haiku-4-5-202510011/5Discharge planning and patient discharge involve complex clinical judgment, coordination with multiple stakeholders (families, social services, specialists), and documented, legally binding decisions tied to individual patient circumstances. Current AI cannot autonomously conduct this end-to-end task; it lacks the clinical authority, legal accountability, and contextual understanding required.
Task automatabilityclaude-sonnet-52/5Discharge planning involves clinical judgment about patient readiness, medication reconciliation, follow-up coordination, and risk assessment that current AI cannot independently perform end-to-end at equal quality.rationale drafts and checklists can be automated but the decision itself requires a physician.
Adoption barriersclaude-haiku-4-5-202510015/5Discharge decisions are the legal and clinical responsibility of a licensed physician; liability, medical board regulations, and healthcare accreditation standards all mandate physician sign-off. Patient safety and informed consent requirements create hard legal barriers to full automation.
Adoption barriersclaude-sonnet-55/5Discharging a patient is a licensed medical act requiring physician authorization and legal responsibility, with high liability for errors, making this a hard-barrier task.
Cost vs. human wageclaude-haiku-4-5-202510012/5AI tools that support discharge planning (documentation, risk flagging) are relatively low-cost, but the human physician oversight cost remains high because discharge decisions are high-stakes and non-delegable. The cost ratio does not approach parity with the full labor of a hospitalist's discharge function.
Cost vs. human wageclaude-sonnet-52/5AI can cheaply draft summaries but a physician must still review, verify clinical status, and sign off, so overall cost savings versus the human's time are only partial.
Technical feasibility todayclaude-haiku-4-5-202510012/5Deployed AI systems can assist with some elements (e.g., generating discharge summaries, flagging readmission risk), but no production system reliably conducts comprehensive discharge planning or executes discharge decisions. Products exist at prototype/pilot stage but with gaps in safety, liability coverage, and regulatory acceptance.
Technical feasibility todayclaude-sonnet-52/5Some hospital systems use AI-assisted discharge summary drafting and readmission risk scoring, but no deployed product independently conducts full discharge planning or authorizes discharge.

Prescribe medications or treatment regimens to hospital inpatients.

13

CI 3–23 · exposure 13 · augmentation 75 · importance 4.9/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510013/5Hospitals are adopting clinical decision-support and AI-assisted diagnostic tools at a moderate pace, but these remain assistive. Autonomous prescription AI is not yet in measurable production deployment; adoption remains pilot-stage in research settings.
Sector adoption velocityclaude-sonnet-52/5Hospitals are adopting AI clinical decision support and documentation tools, but actual prescribing authority automation is essentially nonexistent and highly regulated, slowing adoption for this specific task.
Augmentation potentialclaude-haiku-4-5-202510014/5AI-powered drug interaction checkers, dosing calculators, clinical guideline summaries, and literature retrieval meaningfully assist hospitalists by reducing manual lookup time and flagging safety concerns, thereby raising prescribing accuracy and speed while the physician remains the decision-maker.
Augmentation potentialclaude-sonnet-54/5AI clinical decision support, drug interaction checking, and diagnostic assistance tools meaningfully help physicians make faster, more informed prescribing decisions while they retain full authority and responsibility.
Task automatabilityclaude-haiku-4-5-202510012/5While AI can assist in drug interaction checking, dosage calculations, and guideline-based recommendations, prescribing requires integration of patient history, real-time clinical context, contraindications, and clinical judgment that current AI cannot reliably execute end-to-end without substantial human oversight. The task falls well short of the 50% time-saving bar for full automation.
Task automatabilityclaude-sonnet-51/5Prescribing requires clinical judgment, legal accountability, and physical examination context that current AI cannot autonomously perform end-to-end for inpatients today.'
Adoption barriersclaude-haiku-4-5-202510015/5Prescribing is legally restricted to licensed physicians (MD/DO) in virtually all jurisdictions; only a licensed provider can author and sign medication orders. Liability for adverse events, regulatory oversight (FDA, state boards), and the requirement for real-time clinical judgment create hard legal barriers to autonomous AI substitution.
Adoption barriersclaude-sonnet-55/5Prescribing is tightly regulated, requires licensure (DEA/state medical board), and carries direct liability, making it a hard legal barrier to automation.
Cost vs. human wageclaude-haiku-4-5-202510012/5Current AI tools (decision support, literature search) reduce some clerical and research overhead, but a hospitalist's salary is loaded with benefits, on-call compensation, and malpractice liability that AI integration does not yet undercut. The cost of oversight, validation, and legal/compliance infrastructure offsets savings.
Cost vs. human wageclaude-sonnet-51/5AI cannot legally substitute for the physician's act of prescribing, so there is no valid cost comparison for full task replacement; any AI use still requires physician time and liability coverage.
Technical feasibility todayclaude-haiku-4-5-202510012/5Clinical decision-support systems exist in production (e.g., CPOE systems with alerts), but they function as aides to human prescribers, not autonomous prescribers. No deployed product reliably prescribes treatment regimens independently; all systems require a licensed physician to author and sign the order.
Technical feasibility todayclaude-sonnet-51/5No deployed product autonomously prescribes medications to hospital inpatients without a licensed physician's authorization; decision-support tools only suggest options.

Admit patients for hospital stays.

11

CI 3–20 · exposure 13 · augmentation 75 · importance 4.8/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510012/5Hospital adoption of AI for core admission decisions remains minimal; pilots focus on documentation and workflow support, not autonomous admission. The healthcare sector's regulatory conservatism and liability concerns slow adoption of AI touching core clinical authority.
Sector adoption velocityclaude-sonnet-52/5Healthcare adoption of AI for clinical decision tasks remains slow and heavily regulated, with pilots common but production autonomous decision-making rare.
Augmentation potentialclaude-haiku-4-5-202510014/5AI already assists hospitalists significantly by automating documentation, suggesting appropriate admission criteria, gathering relevant prior records, and summarizing patient histories. These augmentations raise physician efficiency substantially while the clinician retains final decision authority.
Augmentation potentialclaude-sonnet-54/5AI tools can meaningfully assist hospitalists by summarizing patient histories, drafting admission notes, flagging risk factors, and suggesting orders, improving efficiency while the physician retains decision authority.
Task automatabilityclaude-haiku-4-5-202510012/5While AI can assist with data entry, documentation, and triage scoring, patient admission involves complex clinical judgment, legal signing authority, and real-time interaction with patients and families that cannot be fully automated today. Current systems cannot reliably handle the exception cases, consent processes, and clinical decision-making required for safe autonomous admission.
Task automatabilityclaude-sonnet-51/5Admitting a patient requires physical examination, clinical judgment, legal responsibility, and direct patient interaction that current AI cannot perform end-to-end.
Adoption barriersclaude-haiku-4-5-202510015/5Admission requires legal physician authority and signature; licensing and liability laws mandate a licensed physician to evaluate and admit patients. Regulatory frameworks (CMS, state medical boards) explicitly require credentialed clinician responsibility, creating hard legal barriers to automation.
Adoption barriersclaude-sonnet-55/5Admitting a patient is a licensed medical act requiring physician authority, legal accountability, and regulatory compliance (e.g., hospital credentialing, EMTALA), making substitution essentially prohibited.
Cost vs. human wageclaude-haiku-4-5-202510012/5Current AI tools for admission support (documentation, order entry) have moderate costs relative to the high-wage hospitalist labor involved, and still require substantial physician oversight. Full end-to-end cost replacement is not achievable today because human judgment remains irreplaceable.
Cost vs. human wageclaude-sonnet-51/5Since AI cannot perform the core clinical decision and legal act of admission, there is no substitutable AI cost basis; a physician's involvement remains mandatory.
Technical feasibility todayclaude-haiku-4-5-202510012/5No deployed AI system independently admits patients in production settings. Assistive tools exist for documentation and preliminary screening, but the final admission decision and legal attestation remain physician responsibilities. Products are research-stage or limited to narrow administrative subtasks.
Technical feasibility todayclaude-sonnet-51/5No deployed product independently admits patients; AI is at most used for documentation support behind a licensed physician's decision.

Train or supervise medical students, residents, or other health professionals.

9

CI 3–16 · exposure 5 · augmentation 50 · importance 3.8/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510012/5Healthcare institutions have begun experimenting with AI-assisted learning platforms, but actual replacement of clinician trainers remains minimal; adoption is limited by regulatory constraints and the value placed on human mentorship in medical education.
Sector adoption velocityclaude-sonnet-52/5Healthcare training programs are adopting AI slowly for ancillary tasks like question banks or simulation, but supervisory/teaching roles remain untouched by production AI.
Augmentation potentialclaude-haiku-4-5-202510013/5AI can assist hospitalists by generating case summaries, quizzes, or evidence-based teaching points, moderately improving preparation efficiency; however, the core supervisory and mentoring interactions remain human-centered.
Augmentation potentialclaude-sonnet-53/5AI can help generate teaching materials, quiz questions, case summaries, or explain concepts, assisting hospitalists in preparing or supplementing trainee education.
Task automatabilityclaude-haiku-4-5-202510011/5Training and supervising medical professionals requires real-time clinical judgment, mentorship, relationship-building, and adaptive teaching tailored to individual learners' needs—tasks that demand human expertise, accountability, and presence that current AI cannot replicate end-to-end.
Task automatabilityclaude-sonnet-51/5Training and supervising trainees requires live clinical judgment, mentorship, real-time feedback on patient care decisions, and professional accountability that AI cannot perform end-to-end.
Adoption barriersclaude-haiku-4-5-202510014/5Medical licensing boards, accreditation bodies (ACGME, LCME), and institutional credentialing requirements mandate that licensed physicians directly supervise trainees; liability and patient safety law create hard barriers to full AI substitution.
Adoption barriersclaude-sonnet-55/5Medical education and supervision of trainees legally requires licensed, credentialed physicians who bear liability for patient care and trainee sign-off.
Cost vs. human wageclaude-haiku-4-5-202510012/5AI tutoring or content generation tools exist but remain supporting aids; the cost of AI systems plus necessary human oversight does not yet undercut the loaded wage of an experienced hospitalist trainer.
Cost vs. human wageclaude-sonnet-51/5There is no viable AI substitute for supervisory responsibility, so cost comparison favors the human entirely since AI cannot deliver the output at all.
Technical feasibility todayclaude-haiku-4-5-202510012/5While AI can assist with generating educational materials or answering questions, no deployed product reliably performs the full supervisory and training role in production medical settings; human oversight and sign-off remain essential.
Technical feasibility todayclaude-sonnet-51/5No deployed product supervises residents or trains medical students in clinical practice; AI is at most a reference or quiz tool, not a supervisory system.

Diagnose, treat, or provide continuous care to hospital inpatients.

5

CI 3–7 · exposure 5 · augmentation 63 · importance 4.9/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510012/5Healthcare AI adoption remains cautious and heavily regulated. While diagnostic support tools are piloting in some hospitals, actual displacement of hospitalist decision-making authority has been negligible. Adoption is constrained by liability concerns, regulatory scrutiny, and clinician resistance to ceding judgment.
Sector adoption velocityclaude-sonnet-52/5Healthcare adoption of AI for clinical decision-making remains slow and cautious due to regulatory, liability, and safety concerns despite growth in ancillary AI tools.
Augmentation potentialclaude-haiku-4-5-202510013/5AI can meaningfully assist hospitalists through clinical decision support, literature synthesis, documentation automation, and diagnostic imaging interpretation, improving speed and reducing cognitive load on parts of the workflow. However, augmentation is currently limited to narrow tasks within the broader responsibility of continuous care.
Augmentation potentialclaude-sonnet-54/5AI substantially assists hospitalists via clinical decision support, diagnostic suggestion tools, documentation, and literature synthesis, improving efficiency while the physician retains responsibility.
Task automatabilityclaude-haiku-4-5-202510011/5This task requires real-time clinical judgment, continuous physical examination, treatment adjustment, and direct patient interaction that current AI cannot perform end-to-end. While AI can assist with diagnosis support or documentation, the core responsibility of continuous care—adjusting treatments, managing complications, and making life-or-death decisions—remains fundamentally a human-performed task.
Task automatabilityclaude-sonnet-51/5Diagnosing, treating, and providing ongoing inpatient care requires physical examination, real-time clinical judgment, and legal accountability that current AI cannot perform end-to-end.'
Adoption barriersclaude-haiku-4-5-202510015/5Hospitalists must be licensed physicians and are legally responsible for patient diagnosis and treatment decisions. Malpractice liability, regulatory requirements (state medical boards, Joint Commission standards), and direct patient contact mandates create hard barriers to substitution of the full task with automated systems.
Adoption barriersclaude-sonnet-55/5Medical licensure, malpractice liability, and legal requirements mandate that a licensed physician diagnose and direct inpatient treatment, making this a hard-barrier task.
Cost vs. human wageclaude-haiku-4-5-202510011/5AI diagnostic support tools and electronic health record systems are far cheaper than a hospitalist salary, but they do not replace the core task of continuous inpatient care. The cost of AI infrastructure, oversight, and liability management for unsupervised clinical decisions would exceed the loaded cost of a physician in any foreseeable scenario.
Cost vs. human wageclaude-sonnet-51/5AI cannot substitute for the physician's core function, so there is no lower-cost AI alternative for the task itself; costs are additive as a support tool, not a replacement.
Technical feasibility todayclaude-haiku-4-5-202510012/5No deployed product reliably performs the full scope of inpatient hospitalist care. AI diagnostic aids exist (e.g., radiology interpretation, clinical decision support) but these are narrow assistive tools, not end-to-end care systems. Hospital workflows still require a licensed physician to own treatment decisions, physical assessments, and patient management.
Technical feasibility todayclaude-sonnet-51/5No deployed product autonomously diagnoses and manages inpatients; AI is confined to decision-support and documentation aids alongside physicians.

Direct, coordinate, or supervise the patient care activities of nursing or support staff.

1

CI 0–3 · exposure 0 · augmentation 38 · importance 4.6/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510011/5There is no meaningful adoption of AI for supervising patient care staff because the task is legally and ethically bound to licensed clinicians. Healthcare organizations cannot and will not attempt to replace physician supervision with AI systems.
Sector adoption velocityclaude-sonnet-52/5Healthcare delivery settings adopt AI unevenly and cautiously for clinical supervisory functions, with slow integration into hands-on care coordination roles.
Augmentation potentialclaude-haiku-4-5-202510012/5AI could assist with scheduling, documentation, or flagging staff performance metrics, but the core supervisory, motivational, and decision-making aspects remain human. Any augmentation is peripheral to the central coordination function.
Augmentation potentialclaude-sonnet-53/5AI can assist with scheduling, documentation, communication summaries, or flagging care issues, supporting the hospitalist's coordination role without replacing direct supervision.
Task automatabilityclaude-haiku-4-5-202510011/5Directing, coordinating, and supervising patient care staff involves real-time decision-making, interpersonal judgment, conflict resolution, and accountability for patient outcomes. Current AI cannot perform these inherently human supervisory and leadership functions end-to-end.
Task automatabilityclaude-sonnet-51/5Directing and supervising staff in real-time patient care requires physical presence, authority, and dynamic judgment that current AI cannot execute end-to-end.
Adoption barriersclaude-haiku-4-5-202510015/5Hard legal and regulatory barriers exist: a licensed physician must legally supervise patient care activities and be accountable for staff performance. Hospital credentialing, malpractice liability, and patient safety regulations mandate human supervisory authority.
Adoption barriersclaude-sonnet-55/5Clinical supervision of care teams is legally and organizationally tied to licensed physician authority, with strong liability and regulatory requirements for human accountability.
Cost vs. human wageclaude-haiku-4-5-202510011/5AI cannot perform this task, so cost comparison is not applicable. Supervision and coordination remain exclusively in the realm of human staff with attendant labor costs.
Cost vs. human wageclaude-sonnet-51/5There is no AI substitute performing this supervisory role, so cost comparison favors the human by default since no viable AI alternative exists.
Technical feasibility todayclaude-haiku-4-5-202510011/5No deployed product reliably performs the supervisory coordination of patient care teams or substitutes for a licensed physician's responsibility to direct staff. This requires legal accountability and real-time human judgment that AI systems do not currently provide.
Technical feasibility todayclaude-sonnet-51/5No deployed product performs supervisory direction of nursing/support staff in clinical settings; this remains entirely a human management function.

Attend inpatient consultations in areas of specialty.

1

CI 0–3 · exposure 0 · augmentation 50 · importance 4.4/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510011/5Healthcare is a heavily regulated, risk-averse sector where AI adoption for clinical decision-making is slow. Attending consultations is a core physician responsibility that cannot be delegated to machines under current law and medical standards of care.
Sector adoption velocityclaude-sonnet-52/5Hospitals are adopting AI for documentation and diagnostic support, but the core act of attending consultations remains untouched by automation in production settings.
Augmentation potentialclaude-haiku-4-5-202510013/5AI can assist a hospitalist attending consultations by providing real-time clinical decision support, suggesting differential diagnoses, or summarizing relevant evidence, but the physician remains the decision-maker and primary consultant.
Augmentation potentialclaude-sonnet-53/5AI can assist with pulling patient history, summarizing records, suggesting differential diagnoses, and drafting consult notes, improving efficiency while the hospitalist still conducts the encounter.
Task automatabilityclaude-haiku-4-5-202510011/5Attending inpatient consultations requires real-time clinical judgment, patient interaction, physical examination, and dynamic assessment of complex medical conditions. Current AI cannot perform this end-to-end; it cannot be present at the bedside, conduct physical exams, or make the adaptive decisions required during live consultations.
Task automatabilityclaude-sonnet-51/5Attending an inpatient consultation requires physical examination, real-time clinical judgment, and legal accountability that current AI cannot perform end-to-end.
Adoption barriersclaude-haiku-4-5-202510015/5Strong legal and regulatory barriers exist: only a licensed physician can attend and take responsibility for inpatient consultations. Liability, patient safety, informed consent, and medical practice law all require a credentialed human clinician to be the attending.
Adoption barriersclaude-sonnet-55/5Licensed physician credentialing, hospital privileging, malpractice liability, and regulatory requirements mandate a human physician to attend and document consultations.
Cost vs. human wageclaude-haiku-4-5-202510011/5The cost of deploying AI to manage or attend consultations (infrastructure, integration, oversight, liability) would be far higher than the marginal cost of the hospitalist's time, since the physician must remain accountable and present regardless.
Cost vs. human wageclaude-sonnet-51/5Since AI cannot substitute for the physician's presence and legal role, there is no viable AI-only cost comparison; human labor remains required.
Technical feasibility todayclaude-haiku-4-5-202510011/5No deployed AI system can autonomously attend and lead inpatient consultations. While AI can assist with differential diagnosis or literature review, it cannot replace the attending physician's presence, decision-making authority, or responsibility in a consultation setting.
Technical feasibility todayclaude-sonnet-51/5No deployed product independently conducts inpatient specialty consultations; AI is at most a decision-support tool used by the physician who remains present.

Direct the operations of short stay or specialty units.

1

CI 0–3 · exposure 0 · augmentation 38 · importance 3.9/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510011/5This task involves statutory physician leadership requirements in healthcare; adoption of AI autonomous direction is legally and ethically prohibited, with zero sector movement toward such automation.
Sector adoption velocityclaude-sonnet-52/5Healthcare administration adoption of AI is slow due to regulatory, safety, and liability constraints, with pilots focused on documentation rather than unit management.
Augmentation potentialclaude-haiku-4-5-202510012/5AI tools can assist with data analytics, scheduling optimization, and administrative reporting, but human hospitalists remain essential decision-makers; augmentation is limited to narrow support functions rather than transformative productivity gains.
Augmentation potentialclaude-sonnet-53/5AI can assist with dashboards, staffing analytics, and census/flow predictions that support operational decisions, but the directing function itself remains human-led.
Task automatabilityclaude-haiku-4-5-202510011/5Directing unit operations requires real-time decision-making, staff coordination, resource allocation, and strategic planning that fundamentally depends on human judgment, accountability, and adaptive leadership in complex, unpredictable healthcare environments.
Task automatabilityclaude-sonnet-51/5Directing operations of a clinical unit requires real-time management decisions, staff supervision, and adaptive judgment that current AI cannot perform end-to-end.
Adoption barriersclaude-haiku-4-5-202510015/5Hospital licensing, medical malpractice liability, regulatory oversight, and accreditation standards legally require a licensed physician to direct clinical operations, creating hard legal barriers to automation.
Adoption barriersclaude-sonnet-55/5This requires a licensed physician with administrative authority and legal accountability for patient care decisions and staff oversight, a hard regulatory and liability barrier.
Cost vs. human wageclaude-haiku-4-5-202510011/5AI cannot perform this task, so cost comparison is inapplicable; a hospitalist's leadership role commands significant salary and involves irreplaceable accountability that AI cannot assume.
Cost vs. human wageclaude-sonnet-51/5There is no viable AI substitute performing this managerial task, so cost comparison favors the human physician entirely.
Technical feasibility todayclaude-haiku-4-5-202510011/5No deployed AI system can autonomously direct hospital unit operations; this requires legal accountability, licensure, and continuous adaptive decision-making across clinical, financial, and administrative domains that exceed current AI capabilities.
Technical feasibility todayclaude-sonnet-51/5No deployed AI product manages or directs hospital unit operations autonomously; this remains a human administrative and clinical leadership function.

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.