Will AI replace Physician Assistants? AI exposure 13/100

# Physician Assistants

Physician Assistants: low exposure to AI (13/100), with change likely within 7–15 years. Ambient scribes and EHR assistants are taking the documentation burden; examination, diagnosis, procedures and patient relationships stay with the PA.

- Canonical: https://www.careerguard.ai/reports/physician-assistants
- Markdown: https://www.careerguard.ai/reports/physician-assistants/md
- PDF: https://www.careerguard.ai/reports/physician-assistants/pdf
- Exposure: 13/100
- Window: 7-15 years
- Adoption: Medium Adoption
- Revised: 2026-10-05
- Free to read

## Overview

Ambient scribes and EHR assistants are taking the documentation burden; examination, diagnosis, procedures and patient relationships stay with the PA.

**Impact.** Ambient documentation tools such as Nuance DAX Copilot and Abridge listen to the visit and draft the note, and Epic and other EHRs now use AI to summarise charts, draft patient-portal replies and surface relevant history. Clinical reference tools such as UpToDate and OpenEvidence answer questions at the point of care in seconds. This cuts the after-hours charting that has defined the job for many PAs and gives more of the visit back to the patient. History-taking, examination, differential diagnosis, procedures, prescribing decisions and difficult conversations are unchanged and remain the clinician's responsibility.

**Risk.** Low exposure: documentation and reference tasks automate; clinical assessment, procedures and accountable care remain human. The score is low because the measured generative-AI exposure is low: Microsoft Research rates only a small share of the role's tasks as suited to language models, the Anthropic Economic Index records no meaningful observed usage, and the US Bureau of Labor Statistics places the occupation in the low official tier. The medium adoption rating reflects the reality that health systems are deploying ambient scribes and EHR assistants widely, but these remove documentation and inbox work rather than clinical tasks. What automates is note-writing, chart review, coding suggestions, routine patient messages and literature look-up; what stays human is the examination, diagnosis, procedures, prescribing and the relationship, all under professional licence. Over a 7-15 year window, expect PAs to see more patients with less burnout from paperwork, to rely increasingly on AI decision support that they must still verify, and to face continuing strong demand driven by physician shortages. The role expands rather than contracts.

**Sector readiness.** Ambient Documentation Rolling Out Across Health Systems Large health systems have moved ambient scribe tools from pilot to broad deployment and are enabling AI features inside Epic and other EHRs, so many PAs in hospital and large-group settings already use them daily. Smaller practices and rural clinics adopt more slowly because of cost and integration effort. Clinical decision-support AI is more cautious, constrained by validation requirements and liability.

## Where you stand

Position yourself as a clinician who uses ambient documentation and decision support to see patients well and finish on time, rather than as someone buried in charting.

Build procedural and specialty depth - surgery, emergency, dermatology, orthopaedics - where hands-on skill keeps the work firmly human.

Take on roles in clinical informatics or AI governance within your organisation, where clinicians who understand the tools are scarce and needed.

## What this means for you

- **Adopt the scribe, verify the note.** Tools such as DAX Copilot and Abridge draft good notes most of the time and occasionally get things wrong. Read before you sign; the record is yours.
- **Reclaim the visit.** Use the time saved on documentation to look at the patient rather than the screen. It improves care and is what patients remember.
- **Use decision support as a check, not a crutch.** OpenEvidence, UpToDate and EHR suggestions are fast and useful. Keep your own reasoning in front and treat them as a second opinion.
- **Manage the inbox with AI, carefully.** Drafted portal replies save time but need clinical review and a human tone. Set a rule for what you will and will not let the draft say.
- **Go deeper procedurally.** The more your work involves hands-on procedures and complex presentations, the less any software touches it.
- **Get involved in governance.** Health systems need clinicians to evaluate, configure and monitor AI tools. It is a career path and a way to shape how the tools affect your colleagues.

## Drivers of change

- **Ambient clinical documentation.** Nuance DAX Copilot, Abridge and similar tools draft visit notes from the conversation, removing much of the charting that followed each patient.
- **AI inside the EHR.** Epic and other systems summarise charts, draft patient messages and suggest coding, cutting inbox and review time.
- **Point-of-care reference and decision support.** OpenEvidence and UpToDate answer clinical questions quickly and surface guidelines, accelerating but not replacing clinical reasoning.
- **Workforce shortage and demand.** Physician shortages keep PA demand strong, so efficiency gains translate into more patients seen rather than fewer clinicians.
- **Burnout and retention pressure.** Health systems adopt documentation AI partly to reduce after-hours charting and retain clinicians, accelerating deployment.
- **Low measured task applicability.** Occupation-level measures from Microsoft Research and the Anthropic Economic Index find very little of the clinical work suited to language models, and official exposure is rated low.

## Impact by sector

**Large hospital systems and academic centres.** The most advanced setting, with ambient scribes and EHR AI widely deployed and governance structures forming.

**Primary care and outpatient clinics.** High documentation and inbox load makes these the biggest beneficiaries of ambient tools, though smaller practices adopt more slowly.

**Emergency and surgical settings.** Procedural, high-acuity work remains hands-on; AI helps with documentation and triage support rather than the clinical task.

**Rural and community health.** Adoption lags on cost and connectivity, but AI decision support and telehealth extend what a PA can manage with limited physician back-up.

## Skills to build

- **Clinical assessment and diagnostic reasoning.** The core of the role and untouched by automation; keep building it through case volume, feedback and continuing education.
- **Procedural competence.** Suturing, injections, reductions and specialty procedures are hands-on skills that define value in many PA roles.
- **Working with ambient documentation.** Learning to speak the visit clearly for the scribe and to review drafts efficiently turns the tool into real time savings.
- **Critical use of decision support.** Knowing when to trust and when to question AI suggestions protects patients and keeps your judgement sharp.
- **Patient communication.** Explaining, listening and managing difficult conversations is where patients judge their care and where AI contributes least.
- **Clinical informatics literacy.** Understanding how EHR AI features are configured and evaluated opens roles in governance and improvement.

## Tools in use

### Kinds of tool worth knowing

- **OpenEvidence.** AI clinical question-answering tool grounded in medical literature, increasingly used by clinicians for quick reference.
- **Dragon Medical One.** Speech-recognition dictation for clinical documentation, widely used where ambient scribing has not yet been deployed.

### Named tools

- **Nuance DAX Copilot** ([https://www.nuance.com/healthcare/dragon-ai-clinical-solutions/dax-copilot.html](https://www.nuance.com/healthcare/dragon-ai-clinical-solutions/dax-copilot.html)). Ambient clinical documentation that drafts visit notes from the patient conversation inside the EHR workflow.
- **Epic** ([https://www.epic.com](https://www.epic.com)). Dominant hospital EHR with built-in AI for chart summaries, drafted patient messages and documentation support.
- **Abridge** ([https://www.abridge.com](https://www.abridge.com)). Ambient documentation platform deployed by many health systems to generate clinical notes.
- **UpToDate** ([https://www.wolterskluwer.com/en/solutions/uptodate](https://www.wolterskluwer.com/en/solutions/uptodate)). Clinical reference used at the point of care, now adding AI-assisted search and summaries.

## In practice

**Ambient note in primary care.** The PA starts the scribe at the beginning of a visit, conducts the consultation normally, and reviews and signs the drafted note before the next patient. Benefit: Little or no after-hours charting and more attention on the patient during the visit.

**Chart summary before the encounter.** Before seeing a complex patient, the PA reads an AI-generated summary of recent visits, results and medications in the EHR, then verifies key items in the source record. Benefit: Faster preparation and fewer missed details in long histories.

**Drafted portal replies.** The EHR drafts a response to a routine patient message about a normal result; the PA edits it for accuracy and tone and sends it. Benefit: Keeps the inbox manageable without reducing the quality of communication.

## How this role compares

**Medical Transcriptionists** (More exposed). Speech recognition and ambient documentation have automated the core transcription task almost entirely. Work moves to: Quality review, editing and moving into health-information or coding roles.

**Nurse Practitioners** (Different skills, growing). AI eases documentation and offers decision support while demand for independent primary care providers keeps rising. Work moves to: Autonomous patient management, prescribing and relationship-based care.

**Physical Therapists** (Complementary, less exposed). AI assists with exercise plans and documentation, but assessment and hands-on treatment remain entirely physical. Work moves to: Manual therapy, movement assessment and patient coaching.

## Closing judgement

If you are a PA, the most immediate effect of AI is that the note may finally write itself, and that is worth embracing carefully. The tools do not examine the patient, decide what is wrong or take responsibility for the plan; you do. Learn the ambient scribe and EHR assistants well, check what they produce before you sign it, and use the time they free for the clinical and human parts of the visit. With physician shortages continuing, PAs who are efficient, procedurally skilled and good with patients are in a strong position.

## Evidence and revisions

**Revised 5 October 2026.** Score 13; window 7-15 years (unchanged).

Exposure Index v2. Inputs: task applicability 10/100 (Microsoft AI applicability score 0.05 for Physician assistants); observed usage 0/100 (Anthropic observed exposure 0.00); official exposure tier 10/100 (BLS: low); labour-market trajectory not yet mapped for this occupation, so its weight was spread across the other inputs; published adoption rating 40/100 (medium adoption). Weighted base 12.9. Final score 13. New report: the window of 7-15 years is set from the score band.

**How the figure is built (Exposure Index v2).**

| Input | Scaled (0–100) | Weight | Points |
| --- | ---: | ---: | ---: |
| Task applicability (Microsoft Research, AI applicability score) | 10 | 39% | 4.0 |
| Observed usage (Anthropic Economic Index, observed exposure) | 0 | 22% | 0.0 |
| Official exposure tier (US BLS AI-exposure category) | 10 | 22% | 2.2 |
| Labour-market trajectory (US BLS projected employment change 2025–35) | not measured | — | — |
| Published adoption rating (This report’s adoption level) | 40 | 17% | 6.7 |
| **Weighted base** | | | **12.9** |
| **Exposure score** | | | **13** |

### Measures behind the score

- **US Bureau of Labor Statistics, Employment Projections 2025–35 and AI Exposure Categories (27 August 2026).** AI-exposure tier: Low. Projected employment change not yet mapped for this occupation. Matched to Physician assistants. [publisher](https://www.bls.gov/news.release/ecopro.htm) · [PDF](https://www.bls.gov/news.release/pdf/ecopro.pdf) · [archived copy](https://vhjtvcmznlsxakprdimp.supabase.co/storage/v1/object/public/research-public/bls-employment-projections-2025-35.pdf) · [data](https://www.bls.gov/emp/ind-occ-matrix/ai-exposure-categories.xlsx)
- **Microsoft Research, Working with AI: Measuring the Applicability of Generative AI to Occupations (10 July 2025).** AI applicability score 0.05 for SOC 29-1071; scaled to 10/100 as the task-applicability input. [publisher](https://arxiv.org/abs/2507.07935) · [PDF](https://arxiv.org/pdf/2507.07935) · [archived copy](https://vhjtvcmznlsxakprdimp.supabase.co/storage/v1/object/public/research-public/microsoft-working-with-ai-2025.pdf) · [data](https://github.com/microsoft/working-with-ai)
- **Anthropic, Anthropic Economic Index report: Cadences (26 June 2026).** Observed exposure 0.00 for SOC 29-1071; scaled to 0/100 as the observed-usage input. [publisher](https://www.anthropic.com/research/economic-index-june-2026-report) · [PDF](https://cdn.sanity.io/files/4zrzovbb/website/9e0eadc8097864886c5d5060ebb1f89b02ea29d6.pdf) · [data](https://huggingface.co/datasets/Anthropic/EconomicIndex)
- **UK Department for Science, Innovation and Technology, Assessment of AI capabilities and the impact on the UK labour market (28 January 2026).** UK context: around 70% of UK workers are in occupations with tasks AI could perform or enhance, above the US average; a one-standard-deviation rise in exposure was associated with a 3.9% fall in UK job postings. [publisher](https://www.gov.uk/government/publications/assessment-of-ai-capabilities-and-the-impact-on-the-uk-labour-market/assessment-of-ai-capabilities-and-the-impact-on-the-uk-labour-market) · [archived copy](https://vhjtvcmznlsxakprdimp.supabase.co/storage/v1/object/public/research-public/dsit-uk-labour-market-assessment-2026.pdf)

Full research library, with licences and archived copies: https://www.careerguard.ai/sources

## Method and sources

Each report was written from a large body of published research. The exposure score itself is computed, not written: it is the CareerGuard Exposure Index, a weighted average of occupation-level measures from the US Bureau of Labor Statistics (AI-exposure classification and 2025–35 projections), Microsoft Research (AI applicability scores) and Anthropic (observed exposure), together with the adoption rating published on the report. The organisations and publications below are the standing literature behind the narrative sections. Every source, with dates, licences and archived copies where we are permitted to hold them, is catalogued in the research library.

Exposure Index v2 (October 2026). Each input is scaled to 0–100 and weighted: task applicability 35% (Microsoft AI applicability score ÷ 0.5), observed usage 20% (Anthropic observed exposure ÷ 0.75), official exposure tier 20% (BLS very high = 100, high = 70, moderate = 40, low = 10), labour-market trajectory 10% (50 − 2.5 × projected % employment change), published adoption rating 15% (very high = 85, high = 70, medium-high = 55, medium = 40, low-medium = 25, low = 10). Inputs not measured for an occupation are dropped and the remaining weights renormalised. An editorial adjustment of at most ±12 points is allowed only for automation channels the measures cannot see (robotics, self-service, machine vision, medical imaging, RPA/OCR, generative video) and is always logged with its reason. Scores are whole numbers, not rounded to five. The change window shifts one notch (a year at each end) per ten points of movement.

### Global and macroeconomic impact of AI on work

- **World Economic Forum:** The Future of Jobs Report series — Employer survey of expected job growth and decline, skill shifts and technology adoption (2020, 2023 and 2025 editions); Artificial Intelligence and the Future of Entry-Level Work (2026).; AI governance and transformation reports — Frameworks on ethical AI, talent strategy and industry transformation.
- **McKinsey Global Institute:** AI, Automation and the Future of Work series — Research quantifying automation potential by task, sector and demographic, from "Jobs Lost, Jobs Gained" to "Agents, robots, and us" (2025).; Industry-specific reports — Financial services, healthcare, manufacturing and others.
- **PwC:** Global AI Jobs Barometer — Annual analysis of job postings and productivity by AI exposure (2024–2026 editions).; Upskilling Hopes and Fears survey — Employee perceptions and readiness.
- **Microsoft Research and Anthropic:** Working with AI (2025); Anthropic Economic Index (2025–2026) — Occupation-level usage data from Copilot and Claude conversations, the two observed-usage measures behind the 2026 revision.
- **Stanford Digital Economy Lab and Stanford HAI:** Canaries in the Coal Mine? (2025–2026); AI Index Report (annual) — Payroll evidence on early-career employment in exposed occupations; annual measurement of AI capability, investment and adoption.
- **Deloitte:** Human Capital Trends series — Workforce, talent and HR technology trends.; Tech Trends series — Emerging technologies and their business implications.
- **Accenture:** Technology Vision series — Forward-looking analysis of emerging technology, with emphasis on AI.; Fjord Trends — Design, innovation and human experience in a digital world.
- **Boston Consulting Group:** AI/ML insights and industry solutions — "The AI Revolution in the Workplace" and related research.
- **EY:** AI and workforce reports — Adoption, talent strategy and ethics.
- **IBM Institute for Business Value:** AI and automation studies — Business models, workforce evolution and leadership.
- **OECD:** AI Policy Observatory — International data and policy on AI, labour markets and skills.; Employment Outlook — Labour-market trends including technological impact (2023–2026 editions).
- **International Labour Organization:** Generative AI and Jobs: A Refined Global Index of Occupational Exposure (2025) — Task-level exposure gradients for every ISCO occupation; successor to the 2023 global index.
- **International Monetary Fund:** Staff Discussion Notes on AI and work (2024, 2026) — Complementarity framing: where AI augments and where it substitutes.
- **UK Department for Science, Innovation and Technology:** Assessment of AI capabilities and the impact on the UK labour market (2026) — UK occupational exposure and job-posting evidence.
- **Brookings Institution:** AI and automation research — Economic and social implications, displacement and skills.
- **Yale Budget Lab and Goldman Sachs Research:** Tracking the Impact of AI on the Labor Market; AI and the US labour market (2026) — Aggregate labour-market monitoring; macro displacement estimates.
- **Oxford University (Oxford Martin School):** The Future of Employment — Frey & Osborne and subsequent research on susceptibility to automation.
- **MIT Technology Review:** AI & Work — Reporting on AI research and its implications for industries and jobs.
- **Gartner:** Hype Cycle for Artificial Intelligence — Maturity and adoption of AI technologies.; Future of Work reports — Workplace models and talent strategy.
- **U.S. Bureau of Labor Statistics:** Employment Projections 2025–35; AI Exposure Categories; Occupational Outlook Handbook — Ten-year employment projections and, from the 2025 cycle, an AI-exposure tier for every detailed occupation.
- **Indeed Hiring Lab:** AI at Work Report (2025) and posting-market updates — Skill-level transformation estimates and job-posting trends by occupation.

### Core AI and machine-learning research

- **OpenAI:** Research papers, blog and API documentation — Large language models, generative AI, safety and societal impact.
- **Google DeepMind:** Research papers and blog — Reinforcement learning, AI for science, AGI and ethics.
- **Meta AI:** Research papers and blog — Large language models, computer vision, AI for social good.
- **Hugging Face:** Transformers library and model hub — Open-source state-of-the-art NLP models.
- **TensorFlow and PyTorch:** Documentation and community forums — Core frameworks illustrating practical capability.
- **arXiv:** cs.AI, cs.LG, cs.CV, cs.CL — Pre-print research.
- **NeurIPS and ICML:** Conference proceedings — Top-tier academic research.
- **ACM and IEEE:** Journals and proceedings — ACM Computing Surveys; IEEE Transactions on AI.
- **Kaggle:** Datasets and competition solutions — Applied machine learning on real-world problems.
- **The Alan Turing Institute:** Research and reports — Responsible and applied AI.

### Ethical and responsible AI deployment

- **NIST:** AI Risk Management Framework — Voluntary framework for managing AI risk.
- **European Commission:** AI Act — Risk-tiered legal framework for AI.; Ethics Guidelines for Trustworthy AI — Principles for responsible development.
- **Partnership on AI:** Research and best practice — Responsible AI development.
- **AI Now Institute:** Annual reports — Social implications: power, inequality, rights.
- **ACM FAccT:** Proceedings — Fairness, accountability and transparency.
- **Data & Society:** Publications — Social implications of data-centric technology.
- **WIPO:** Conversation on IP and AI — Intellectual-property implications of AI.
- **IEEE Global Initiative on Ethics of A/IS:** Ethically Aligned Design — Recommendations for ethical AI design.
- **Center for AI and Digital Policy:** Policy briefs — Accountable AI policy.