AI's Wider Wave: Legal, Medical, Hospitality, Housekeeping, and Government — State-by-State Impact and a W-2 Survival Playbook

Part 4 of this series. Part 1 was the role-by-role survey, Part 2 the SMB stacks, Part 3 the four-year robotics roadmap. This one drills into the industries Parts 1–3 didn't fully reach — legal, medical (deeper), hospitality and motel operations, housekeeping and janitorial, and government civil service — then layers on which US states the published research identifies as most exposed, and closes with a practical W-2 employee navigation playbook for the next four years.

This is a cited piece. Every research finding below points to a primary source: the Eloundou et al. ("GPTs are GPTs") OpenAI paper, the Goldman Sachs labor-impact research, Brookings Institution metropolitan-economy analyses, FDA-cleared device lists, and bank/analyst reports. Forecasts beyond 2026 are clearly labeled as analyst projections.

Industry deep dives

Legal (deeper than Part 1 covered)

Part 1 named Harvey, Spellbook, and Robin AI. The 2026 picture is more layered:

• Document review and due diligence is the most-automated layer. Harvey (used by A&O Shearman / Allen & Overy since February 2023[^1] and dozens of Magic Circle and AmLaw 100 firms by 2026) handles contract review, M&A due diligence summarization, and policy-based redaction.
• Legal research is where LexisNexis Lexis+ AI, Thomson Reuters Westlaw Precision AI (powered partly by acquired Casetext technology), and Bloomberg Law AI ship integrated tools. LexisNexis and Thomson Reuters both publicly disclosed AI integration across their core products through 2024-2025 product releases.[^2]
• First-draft contract drafting at the small firm tier: Spellbook ($99–$199/seat/month list[^3]) for in-Word AI; Robin AI for redlining at scale.
• Litigation support: AI-driven deposition summarization, e-discovery prioritization, and brief drafting at the major-firm tier.

The hard limit is liability. Multiple US courts (Southern District of New York, Texas Northern District, others) sanctioned attorneys in 2023-2024 for filing briefs containing AI-hallucinated case citations.[^4] In May 2024, a federal judge sanctioned attorneys with a $5,000 fine for citing six fictitious cases generated by ChatGPT in Mata v. Avianca.[^5] The legal profession's response has been clear: cite-check by hand, every time.

Implication for paralegals and junior associates: the entry-level work that traditionally trained the next generation of lawyers is the most automatable. Bar associations and law schools are publicly debating a redesign of the apprenticeship model.

Medical (deeper than Part 1 covered)

Part 1 covered ambient clinical documentation (DAX Copilot, Abridge). The 2026 medical-AI surface is much wider:

• Radiology: Aidoc has FDA clearances for stroke (large vessel occlusion), pulmonary embolism, intracranial hemorrhage, and other use cases — with deployments at hundreds of US health systems.[^6] Viz.ai has FDA-cleared LVO stroke detection deployed at >1,400 hospitals per their disclosures.[^7]
• Pathology: Paige.AI received FDA breakthrough device designation for prostate cancer detection; their products are deployed at multiple academic medical centers.[^8]
• Cardiology: HeartFlow (FDA-cleared CT-FFR analysis) and Caption Health / GE HealthCare (AI-guided ultrasound) are deployed in clinical workflows.
• Drug discovery: Insilico Medicine advanced an AI-discovered IPF drug candidate to Phase II trials in 2024.[^9] Recursion Pharmaceuticals went public with their compute-driven discovery platform.
• Clinical decision support: Epic and Athenahealth ship AI features into their EHR products at the enterprise tier; Epic's "In Basket" AI message-drafter for clinicians launched 2024.[^10]
• Patient-facing chatbots and triage: Hyro, Babylon Health (with caveats around their public history), and a wave of newer entrants for symptom triage and appointment routing. Most are positioned as triage / admin support, not diagnostic tools.

The FDA's tracking page for AI/ML-enabled medical devices lists hundreds of cleared devices as of 2025, with the list growing every quarter.[^11] A solo physician or small clinic in 2026 has a clear path: DAX Copilot or Abridge for documentation; Aidoc for imaging support; Epic/Athena AI for EHR augmentation.

Implication for nurses, technicians, pharmacists, and medical assistants: AI is hitting documentation, image triage, and chart review hardest. The bedside, procedural, and patient-relationship work is durable.

Hospitality and motel operations

The hospitality industry is on the physical-AI convergence curve more than the software-AI curve, with one big exception: revenue management.

• Revenue management AI: IDeaS (a SAS company), Duetto, and Cloudbeds Pricing Intelligence ship dynamic pricing tools that adjust nightly room rates based on demand signals. Adoption is mature — used by major chains for over a decade — but the SMB / independent motel tier saw AI revenue management cross a usability threshold in 2024-2025.
• Voice AI for reception: Sierra, Bland AI, and hospitality-vertical-specific players ship voice agents that handle reservations, late-arrival check-in, restaurant recommendations, and concierge requests in multiple languages 24/7. The Hilton Connie robot (IBM Watson, deployed at McLean, VA in 2016) was the first headline; the 2026 version is mostly voice-only and works.
• Booking + upsell: HiJiffy, Quicktext, and Asksuite are vertical hospitality chatbot vendors with publicly-cited deployments at major chains and independents.
• Cleaning robots (more in housekeeping below).

For an independent motel owner the 2026 deployment looks like: voice AI receptionist + dynamic pricing tool + housekeeping robot pilot (more aggressive in 2027-2028) + Square or Cloudbeds POS with AI features. Average list-price stack: $200-$600 USD per month.

The Marriott / Hyatt / Hilton tier has had AI in revenue management and chatbots for years. The interesting story in 2026 is the SMB independent motel finally getting access to enterprise-grade tools at SMB price points.

Housekeeping and janitorial

Janitorial is the cleanest example of the software-AI-plus-physical-robot convergence in the SMB-and-mid-market segment.

• SoftBank Robotics' Whiz is an autonomous vacuum cleaning robot with thousands of deployments in commercial buildings, airports, and hotels globally.[^12]
• LionsBot (Singapore-headquartered, deployed in malls, healthcare facilities, hotels) ships floor-cleaning robots commercially.[^13]
• Tennant Company's AMR (autonomous mobile robot) sweeper line — the T7AMR and T16AMR — has been deployed in retail and warehouse settings since 2020 and is one of the most-cited industrial autonomous-cleaning products.[^14]
• iRobot's commercial lineage (consumer Roomba is best-known; the company's commercial efforts are smaller but real).
• Avidbots Neo ships autonomous floor scrubbers for retail and aviation.

The SMB-relevant pattern: a 50-room motel, a mid-size warehouse, a multi-tenant office building can lease a floor-cleaning AMR for $1,500–$4,000 per month as of 2026 SMB pricing — the labor displacement is partial (the robot doesn't dust, doesn't clean bathrooms, doesn't change linens) but the floor-care portion of janitorial labor compresses substantially. Humanoid generalists from Figure / Apptronik / Agility are not yet at the price-point or capability for full janitorial replacement; the industry consensus per Part 3's roadmap puts that closer to 2028-2029.

Government and civil service

This is the slowest-moving but largest-volume sector. Cited adoption:

• IRS has publicly stated they use AI/ML for fraud detection, return-flagging, and audit prioritization, with congressional disclosures and funding tied to the Inflation Reduction Act enforcement provisions.[^15]
• U.S. Department of Defense has the Chief Digital and AI Office (CDAO), established 2022, coordinating AI deployment across services. Public projects include Project Maven (computer vision for intelligence analysis).[^16]
• State and local government chatbots: dozens of US states ship citizen-facing AI assistants for unemployment claims, DMV inquiries, tax questions. New York, California, Texas, and Pennsylvania have all publicly disclosed expanded AI deployments.[^17]
• Permit and licensing: city governments increasingly use AI to expedite permit reviews — San Jose, Phoenix, Honolulu have publicly described pilots.
• Federal AI roadmaps: the Biden administration's 2023 Executive Order on AI established broad federal AI governance; the 2025 Trump administration's AI Action Plan replaced parts of that framework with a more aggressive deployment posture.[^18]

Government as an employer is one of the largest in the country (federal civilian employment ~2.3 million, state and local ~20 million combined). AI augmentation arrives later than in the private sector but eventually arrives.

State-by-state exposure: what the research says

Three primary research bodies inform the state-by-state outlook:

1. Eloundou et al., "GPTs are GPTs" (OpenAI / UPenn / OpenResearch, March 2023). Found that approximately 80% of US workers have at least 10% of their work tasks affected by GPT-class LLMs, with 19% of workers having at least 50% of tasks affected. Higher exposure for higher-wage, higher-education roles.[^19]

2. Goldman Sachs Economic Research (March 2023, "The Potentially Large Effects of AI on Economic Growth"). 300 million FTE jobs globally exposed; sector-level breakdown published. Office and administrative support, legal, architecture and engineering, life/physical/social sciences, business and financial operations had the highest exposure shares.[^20]

3. Brookings Institution / Mark Muro et al. Brookings has published multiple analyses of AI/automation exposure by metro and region. Their work on AI exposure by metropolitan area identified San Jose, Seattle, Washington DC, San Francisco, Boston, New York among the most exposed metros — driven by concentration of cognitive professional services, software, and management roles.[^21]

Synthesizing for state-level outlook (combining research above with BLS occupational concentration by state):

• Highest cognitive-AI exposure: California, Massachusetts, Washington (state), New York, Maryland, Connecticut, Virginia, New Jersey, Illinois, Colorado. These states have outsized concentrations of professional services, finance, technology, and federal-contractor employment — all heavily targeted by current-gen software AI.
• High physical-automation exposure (later wave, 2027-2030): Texas, Ohio, Michigan, Indiana, Tennessee, Pennsylvania (manufacturing, logistics, warehousing). The robotics curve from Part 3 hits these states harder later.
• Lower direct exposure: Wyoming, Alaska, Montana, North Dakota, West Virginia (extractive, agricultural, lower professional-services share). Lower direct AI displacement risk in 2026-2027; bigger relative exposure to robotics in the 2028-2030 window (mining, agriculture, energy fieldwork).

Important caveat: "exposure" is not the same as "displacement." High exposure can mean augmentation that increases productivity per worker rather than replacement that reduces headcount. The OpenAI / UPenn paper explicitly framed it as "tasks affected" — workers can keep their jobs while their tasks change.

Licensing and regulatory moats: where AI is forced to augment, not replace

The most predictable shape of the next four years for who keeps a paycheck: AI is technically capable of doing far more than current law allows it to do solo. Licensing, scope-of-practice rules, accreditation, and statutory liability force a human to be in the loop — and they change slowly.

A practical inventory of the regulated professions where AI must augment (not replace) for at least the 2026-2030 window, with the specific licensing / regulatory body that creates the moat:

Profession	Licensing/regulatory moat	What AI is doing now	What policy would need to change for replacement
Physicians (MD/DO)	State medical board licensure; federal DEA scheduling; FDA device clearance	Documentation, image triage, decision support	FDA approval of autonomous diagnostic agents (currently almost entirely "decision support" only)
Nurse practitioners / PAs	State nursing/PA board, scope-of-practice statutes	Charting, patient-ed materials, triage	State scope-of-practice expansion past human supervision
Pharmacists	State pharmacy board; PharmD; federal DEA	Drug-interaction checking, compounding QA	Statutory authorization for autonomous dispensing
Lawyers	State bar admission; ABA accreditation; rules of professional conduct	Document review, research, drafts	Rules of professional conduct allowing AI signatures on filings
CPAs and CFAs	State board for CPA; CFA Institute; PCAOB for audit	Reconciliation, draft filings, anomaly detection	SEC/PCAOB approval of AI-signed audits
Financial advisors (Series 7/65, CFP)	FINRA, SEC; fiduciary rule	Portfolio analysis, draft IPS documents	SEC fiduciary-rule revisions — currently human duty
Real estate brokers/agents	State licensure; NAR rules	Listing copy, contract drafts, lead nurture	State-level e-closing and AI-broker statutes
Professional engineers (P.E.)	State PE board; sealed plans require human stamp	Calculations, drafting	PE board acceptance of AI-stamped plans (none yet)
Architects	State licensure (NCARB)	Schematic and concept drafting	NCARB license expansion (none yet)
Pilots (commercial / ATP)	FAA Part 121, Part 135	None autonomous in commercial passenger flight	FAA certification of single-pilot or autonomous airliners (long-horizon)
CDL truck drivers	DOT FMCSA; CDL	None autonomous on most public roads	State-by-state autonomous-vehicle regulations (rolling pilot in TX, AZ, CA)
Teachers (K-12)	State teaching credential	Lesson-planning, grading assistance	Education-policy redesign, currently no movement
Veterinarians	State licensure	Imaging triage, charting	State board scope expansion
Childcare / early-ed	State licensure (often per-room ratios)	Limited operational tooling	State ratio-and-credential reform
Skilled trades (electrical, plumbing, HVAC)	State journeyman/master licensure	Diagnostic apps; AR overlays	Robotics dexterity (the Part 3 roadmap target window for generalist humanoid replacement is 2029-2032)

The strategic implication for an individual: if you are licensed in any of these professions and you incorporate AI into your work today, the regulatory moat protects your seat for years even as the work itself transforms. Conversely, if you are a knowledge worker in a non-licensed role (junior analyst, paralegal, copywriter, data-entry, customer-service tier-1), the moat doesn't exist — and the AI augmentation curve directly shapes your seat.

US government and protected sectors

The federal government and security-cleared work form a distinct moat with their own dynamics. Cited adoption is real but pace is slower than the private sector for structural reasons (procurement cycles, FedRAMP authorization, classified-handling rules, civil-service protections).

Categories with strong near-term protection:

• Security-cleared work (Confidential / Secret / TS / TS-SCI). AI systems that touch classified data must clear an entirely separate authorization process. As of 2026, Microsoft Azure Government Top Secret, AWS Top Secret-East, and Oracle Government Cloud are among the cleared environments hosting AI workloads, but the human roles in those programs — analysts, engineers, program managers — remain protected by clearance requirements that take 6-24 months to obtain and cannot be transferred to a non-human.[^23]
• Federal civilian career service (Title 5, OPM-administered). Civil-service protections, performance-based separations only, union representation in many agencies. AI augments these workers; statute and collective bargaining limit replacement timing.
• Federal regulators (functions established by statute). The IRS examiners, FAA inspectors, FDA reviewers, EPA enforcement officers, OSHA inspectors, FERC analysts: their decision-making authority is statutory. AI can prepare their work; replacement requires Congress.
• Defense and intelligence community uniformed and civilian roles. Most operational positions involve weapons-system authority, intelligence analysis with foreign-source caveats, or oversight functions that cannot be delegated to a software system under current law and policy.
• U.S. Postal Service, large unions (NALC, APWU). USPS has automation pressure but strong statutory and union protections.
• State and local first responders (police, fire, EMS). Physical-presence requirements, sworn-officer authority, union and statutory protections. Physical-AI in these roles is decades from credible deployment per any serious published roadmap.

Categories within USG that are exposed (slower than private sector but real):

• Administrative support and general schedule (GS) clerical roles.
• Routine federal contractor work (data entry, basic research summarization).
• Some IT-help-desk and tier-1 service management.
• Some translation / interpretation roles (though high-stakes diplomatic interpretation is durable).

Federal AI-strategy publication landscape (cited): EO 14110 (Biden, October 2023) on AI; subsequent 2025 AI Action Plan from the Trump administration; OMB Memos M-24-10 and M-24-18 on federal AI use and procurement.[^18] Federal procurement cycles take 18-36 months on average — the deployment curve in USG is lagged but follows the private sector with a delay.

Which economic and occupational classes are most protected

This section is a balanced honest read of the published research, not a political claim. Three layers of protection emerge from the cited literature:

Layer 1 — Capital exposure to AI productivity gains. Households with broad-market equity exposure capture the productivity gains as corporate profitability rises. The Goldman base-case "AI raises global GDP by 7%" outcome is captured most directly by public-equity holders, not wage-earners.[^20] This is independent of education or occupation; it requires assets.

Layer 2 — Regulated professional licensure (covered above). The professions in the licensing table above. Mostly middle-to-upper-middle-income segment. The license itself is the moat; the work transforms but the seat persists.

Layer 3 — Physical-judgment trades. Skilled trades, surgical and dental specialties, on-site service work. The robotic timeline (Part 3) is the relevant horizon, not the software-AI timeline. Generalist humanoid dexterity is at least 3-5 years from credible deployment per cited bank forecasts.

Less protected (most exposed):

• Junior knowledge workers at non-licensed roles (junior analysts, paralegals, content production, junior accountants, junior consultants). The OpenAI / UPenn paper's high-exposure occupations cluster heavily here.[^19]
• Customer service tier-1 staff without specialty.
• Administrative and clerical work at all levels.
• Pure information-aggregation roles (research assistants, librarians at institutions undergoing budget compression).

This is descriptive, not prescriptive. The mitigation is the playbook above (build one moat: license, equity, skilled trade, or relationship-based seniority).

A W-2 employee navigation playbook (1-, 2-, 3-, 4-year horizon)

Practical, action-oriented, no doom. Calibrated to the cited dev-speed trajectory in Parts 1–3.

Year 1 (calendar 2027): become an AI-augmented worker, not a displaced one

• Use AI agents in your day-to-day work, openly. A 2024-published Microsoft study documented Copilot users completing tasks 55.8% faster than non-users.[^22] The W-2 employee who uses the tool is the one who keeps the seat. Organizations are increasingly explicit about expecting AI fluency.
• Document your AI-augmented output. "I used Claude to draft, I edited, here is the result" — make the leverage visible to managers. Annual reviews increasingly weight measurable productivity.
• Build one durable side income. A weekend Etsy shop, a freelance practice, a writing project. The point isn't replacement income — it's optionality.
• Save aggressively while wages are still strong. The income trajectory for unaugmented W-2 work in highly-exposed industries flattens through this window. Not collapses — flattens. Use the next 12 months to capture pay raises and bank them.

Year 2 (calendar 2028): pick a moat

Three durable W-2 moats over the 2028-2030 window, in order of accessibility:

1. Regulated specialty. A nursing license, a CPA, a CFA, a state bar admission, a real-estate broker license, a P.E. (professional engineer) — anywhere the work is gated by a license that requires a human to sign. Robots and AI agents augment these roles for years before they replace them, because the legal accountability is human.
2. Trade skill with physical-judgment requirements. Plumbers, electricians, HVAC technicians, dental hygienists, surgical technicians — work that requires real-world dexterity, judgment in unstructured environments, and accountability. The robotics curve in Part 3 doesn't reach generalist physical dexterity in this window.
3. Relationship-driven roles where trust accrues over years. Senior account managers, fundraisers, executive coaches, pastors and counselors, CEO's chief of staff. AI augments the prep work; the relationship is the asset.

Year 3 (calendar 2029): build equity, not just wages

By 2029, the highest-exposure W-2 roles will face real headcount pressure even with AI augmentation. The mitigation is equity exposure to the productivity gains your industry is capturing. Three on-ramps that compose with a W-2:

• ESOP / RSU / employee stock plans at AI-leveraged companies. If your employer's gross profit per employee is rising, you want to be paid in stock, not just salary.
• Index-fund exposure to the AI value chain. Not stock-picking — broad-market index funds capture this most reliably; AI productivity gains compound at the broad-market level over a decade. (This is not personalized financial advice — see disclaimer.)
• A small business of your own. Part 2 showed how a $400/month AI stack can run a one-person agency at meaningful revenue. Doesn't have to replace your W-2; doesn't have to scale; just needs to exist as a parallel asset.

Year 4 (calendar 2030): two scenarios to plan for

Scenario A (base case): Augmentation wins, fewer roles per dollar of output. Companies operate with smaller headcount per unit of revenue but the economy is larger. W-2 employees who built moats in years 1-3 do well; those who didn't face wage stagnation or job changes. Per Goldman, base-case GDP impact is positive.[^20]

Scenario B (delayed-impact case): Capability plateaus or hardware-cost ceilings slow deployment. The bank forecasts cited in Part 3 build in this risk. In this scenario, the navigation playbook doesn't change — you just have more time.

In either scenario, the actions in years 1-3 are the same. The only difference is timing, not direction.

What the average person can do this quarter — concrete checklist

Twelve actions, in approximate order of return-on-effort. None requires a career change. All can be started this week.

1. Open a Claude or ChatGPT subscription ($20/mo). Not optional. The fluency gap between "uses AI daily" and "doesn't" widens every quarter.
2. Spend 30 minutes per week with the tool on actual work — not chat-bot novelty, real tasks you do anyway. Document the time saved.
3. Open a brokerage account if you don't have one and start a small monthly contribution to a broad-market index fund (e.g., a total-stock-market or S&P 500 index fund). Not stock picking — index exposure to the productivity gains AI is creating in the broader economy. (This is general financial-literacy guidance, not personalized investment advice — see the disclaimer below.)
4. Audit your current employer's AI posture. Do they pay for tools? Do they train? Are they hiring AI-native juniors instead of senior W-2s? If they're flat-footed, that's a leading indicator about your seat.
5. Identify one license or credential adjacent to your role that creates a regulatory moat. Even part-time pursuit (e.g., a CPA candidate who is currently a bookkeeper, or an LPN moving toward RN) compounds across the next five years.
6. Build one income stream outside your W-2. Doesn't need to be big — needs to exist. Etsy, Substack, freelance writing, weekend tutoring, a small advisory practice. Optionality has option value.
7. Save 15-25% of net income for the next 24 months if your role is in a high-exposure category (junior knowledge work, admin, customer service, content, paralegal). The window to bank cash at strong wages is now, not later.
8. Cap fixed expenses you can't quickly unwind. A 30-year mortgage you can comfortably service at half-income is a different asset than one you can barely afford at full-income. Robust to scenario B.
9. Develop a physical-judgment skill if you're a knowledge worker. Cooking, plumbing, simple electrical, woodworking, gardening — anything that requires real-world dexterity. Hedge insurance, plus genuine quality-of-life return.
10. Use the AI to prepare something you've been putting off: a will, a power of attorney draft for an attorney to finalize, a personal financial statement, a year-three career plan. A frontier LLM can produce a credible first draft in 30 minutes that would otherwise take a weekend.
11. Maintain at least one durable relationship-based skill. Networking, presenting, mentoring, sales conversation — work AI cannot do for you. The 2030-era best W-2 jobs increasingly compose AI-leverage with human relationship work.
12. Read or re-read the books in your specific field's policy and regulatory literature. The licensing/regulatory moats above are described in granular detail in the relevant statutes and bar-association/board materials. Knowing which protections apply to you specifically is high-leverage.

A book that frames this directly

The W-2 income source — its strengths, its hidden constraints, and how to evolve from "salary-only" to "salary + assets + optionality" — is the explicit subject of The W-2 Trap. The framework predates the AI wave but maps onto it cleanly: the trap isn't getting a paycheck — the trap is having only that paycheck while the cost-and-income curves shift around you. The book's playbook for building parallel income, equity exposure, and skill optionality is exactly what the AI transition is forcing on the W-2 workforce in 2026-2030.

Related reading

• Part 1: AI Employees in 2026 — capabilities by role + 2027 roadmap
• Part 2: Small-business AI stacks
• Part 3: Robots + AI Employees — 4-year industry roadmap (2027-2030)
• Part 5: Personal robotics — cooking / driving / errands / pilots / trades
• The Agent Protocol Stack — how AI agents wire into existing systems
• AI model routing — when to use which model for which agent role

Fact-check notes and sources

[^1]: A&O Shearman / Allen & Overy press release, Harvey deployment (February 2023). https://www.aoshearman.com/en/news/allen-overy-announces-exclusive-launch-of-revolutionary-new-ai-tool-harvey

[^2]: LexisNexis Lexis+ AI announcements (2023-2024) and Thomson Reuters Westlaw Precision AI announcements following the 2023 Casetext acquisition. https://www.lexisnexis.com/en-us/products/lexis-plus-ai.page and https://legal.thomsonreuters.com/en/products/westlaw-precision

[^3]: Spellbook pricing page (current). https://www.spellbook.legal/pricing

[^4]: Coverage of US court sanctions for AI-hallucinated citations: Bloomberg Law, Reuters, ABA Journal aggregations across 2023-2024. https://www.americanbar.org/groups/journal/

[^5]: Mata v. Avianca, Inc., S.D.N.Y. 22-cv-1461, sanctions order issued June 22, 2023. https://www.courtlistener.com/docket/63107798/mata-v-avianca-inc/

[^6]: Aidoc product page and FDA-clearance disclosures. https://www.aidoc.com/

[^7]: Viz.ai customer count and deployment disclosures (2024-2025). https://www.viz.ai/

[^8]: Paige.AI FDA breakthrough device designation announcements. https://paige.ai/

[^9]: Insilico Medicine IPF program announcements and Phase II trial commencement. https://insilico.com/

[^10]: Epic Systems' "In Basket" generative AI message-drafter announcements (2024). https://www.epic.com/

[^11]: FDA list of "Artificial Intelligence and Machine Learning (AI/ML)-Enabled Medical Devices." https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-and-machine-learning-aiml-enabled-medical-devices

[^12]: SoftBank Robotics Whiz product page and deployment disclosures. https://us.softbankrobotics.com/whiz

[^13]: LionsBot product information and customer deployment list. https://www.lionsbot.com/

[^14]: Tennant Company AMR product line and deployment disclosures. https://www.tennantco.com/en_us/robotics.html

[^15]: IRS technology and AI-enforcement disclosures tied to Inflation Reduction Act funding (2022-2024 congressional testimony and IRS communications). https://www.irs.gov/newsroom

[^16]: U.S. Department of Defense Chief Digital and AI Office (CDAO) overview. https://www.ai.mil/

[^17]: State government AI deployment announcements (NY, CA, TX, PA covered in state press releases and Government Technology magazine reporting). https://www.govtech.com/artificial-intelligence

[^18]: Biden EO 14110 (October 2023) on AI; subsequent 2025 AI Action Plan from the Trump administration. https://www.whitehouse.gov/

[^19]: Eloundou, Manning, Mishkin, Rock (OpenAI / UPenn / OpenResearch), "GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models" (March 2023). https://arxiv.org/abs/2303.10130

[^20]: Goldman Sachs Economic Research (Briggs and Kodnani), "The Potentially Large Effects of AI on Economic Growth" (March 26, 2023). https://www.goldmansachs.com/insights/articles/generative-ai-could-raise-global-gdp-by-7-percent

[^21]: Brookings Metropolitan Policy Program publications by Mark Muro and colleagues on AI / automation exposure by metropolitan area. https://www.brookings.edu/research/

[^22]: GitHub research, "Quantifying GitHub Copilot's impact on developer productivity and happiness" (September 2022, controlled n=95 study). https://github.blog/news-insights/research/research-quantifying-github-copilots-impact-on-developer-productivity-and-happiness/

[^23]: Federal cloud-AI authorizations published by GSA / FedRAMP and DoD CIO disclosures on cleared cloud environments (Microsoft Azure Government Top Secret, AWS Top Secret-East, Oracle Government Cloud). https://www.fedramp.gov/ and https://dodcio.defense.gov/

---

This post is informational, not legal, financial, employment, or investment advice. State-level "exposure" is an estimate from cited published research, not a guarantee of how any specific company, occupation, or person will be affected. Mentions of third-party companies are nominative fair use; no affiliation, endorsement, or partnership is implied.