Epic’s AI Scribe Saves Clinicians Up to 60 Minutes Daily, Group Health Reports

The Launch: What Epic Actually Shipped

Epic spent years watching more than 120 AI scribe startups (HealthExec found) raise billions and colonize its customers' workflows. In August 2025, it finally answered the question every health system CIO had been asking: when does the EHR itself just do this?

The move intensifies competition for AI scribe startups, forces health systems to reconsider vendor timelines, and tightens the market for clinical AI talent. Epic's native ambient documentation tool, branded "Art," arrives pre-integrated, bundled into existing Epic licenses, and backed by the largest clinical dataset in U.S. healthcare. That combination rewrites the baseline for every vendor, buyer, and engineer in the space.

The answer arrived at Epic's Users Group Meeting on August 20, 2025. After a leak to Politico on August 7 forced the company's hand, Epic confirmed it had built a native ambient AI scribe powered by Microsoft's Dragon Ambient AI (the former Nuance DAX engine) and connected to Cosmos, Epic's de-identified clinical database spanning more than 1,760 hospitals and 300 million patients (CNBC found). The company also revealed a parallel partnership with Abridge, the $5.3 billion valuation (Precedence Research reported) startup whose technology already runs at Johns Hopkins, Mayo Clinic, and Kaiser Permanente. Both integrations are native now: no bolt-on middleware, no separate login, no API handshake that breaks every quarter.

Art is the clinician-facing half. Epic describes it as an "active AI digital colleague" that drafts clinical notes, anticipates information needs (pulling blood-pressure trends, updating family history, queuing orders) and sits inside the same EHR interface physicians already live in. Penny handles the revenue cycle: generating appeal letters for denied claims and surfacing coding suggestions, two features already live. Emmie faces patients inside MyChart, answering lab-result questions, proposing appointment slots, and flagging overdue screenings. MyChart Central, also announced at UGM, lets patients log in once across every Epic health system they visit. Behind all of it, Epic is training proprietary foundation models (Cosmos AI) on the 8 billion encounters Cosmos has ingested so far (CNBC's figures put). "We're just getting started," Seth Hain, Epic's senior vice president of research and development, told the UGM crowd.

The rollout is real. Group Health Cooperative of South Central Wisconsin went live first, reporting up to 60 minutes saved per clinician per day (Distilinfo reported), improved provider well-being, and better patient experience scores. Mount Nittany Health in State College, Pennsylvania, planned to activate Art when it launched Epic in May 2026. Penn Medicine was evaluating; UT Austin would use it for ambulatory care starting April 2026, with inpatient following in 2030. Epic has not published pricing, but Diane Hunt, MD, Mount Nittany's chief health information officer, expects it to land near Microsoft Dragon Copilot's pricing. She estimates that cost at "a couple hundred dollars per month per provider" (according to Distilinfo).

Epic launched AI Charting in early February 2025 (Distilinfo's data shows). Epic says it's developing roughly 200 different artificial intelligence features (CNBC reported) for patients, clinicians and insurers. Art, Penny, Emmie, and Cosmos AI are the first wave that ships inside the EHR rather than beside it.

Why Epic's Moat Just Got Deeper

Epic's competitive position was already formidable before the AI scribe launch. KLAS Research data from May 2026 shows the company commands 43.7% of the U.S. acute care EHR market, up from 42.3% a year earlier, and controls 56.9% of hospital beds. That translates to roughly 3,300 hospitals, 71,000 clinics, and 325 million patient records worldwide. Oracle Health, the nearest rival, sits at 21.9% of hospitals and 20.4% of beds after losing a net 56 hospitals and 14,676 beds in 2025 alone. Since 2021, Epic has added a net 568 hospitals while Oracle has lost 173. The gap isn't narrowing; it's widening.

The structural advantages run deeper than market share. Epic remains privately held under founder Judy Faulkner, who at 82 still runs the company from Wisconsin and has never sold her voting shares. A trust structure she designed will keep Epic private after her death, governed by a voting committee of family members and longtime employees, overseen by a "Trust Protector Committee" of three health-care leaders who can sue if the rules aren't followed. This insulation from quarterly-pressure capital markets means Epic can invest in decade-long product cycles, like the April 2023 Microsoft Azure OpenAI partnership, without justifying each quarter's ROI to public shareholders. Faulkner's aversion to acquisition offers is legendary: former GE CEO Jeff Immelt once tried to buy Epic and got a five-minute meeting. "Others have asked to come and persuade us," Faulkner told CNBC, "and I've heard our staff say to them, 'Just leave your car running.'"

Customer lock-in operates at multiple layers. Each health system gets a dedicated "BFF" (best friend forever), a named Epic staffer available 24/7 with no third-party call centers. Faulkner herself answers CEO emails within hours and holds recurring video calls with senior executives. When Ardent Health CEO Marty Bonick debated a conversion, Faulkner flew in, presented for 90 minutes, and "was not watching the clock." That partnership model shows up in KLAS surveys: buyers cite "platform strength, partnership, and standardization" as primary decision drivers, alongside regional data exchange. Epic's Community Connect program extends the same platform to smaller practices and critical-access hospitals, with 49 such wins in 2025 alone, seven under 200 beds, spreading the moat downstream without independent implementation costs.

Switching costs are brutal. Epic's Cosmos platform, launched early 2024, aggregates de-identified data across the network for real-time clinical insights. Care Everywhere and EpicCare Link handle interoperability within the Epic ecosystem and with affiliates. Health systems that have spent years customizing workflows, building order sets, and training clinicians on Epic's interface face substantial retraining costs and workflow disruption. The 2026 KLAS report notes that overall EHR buying activity fell roughly 40% from 2024 and nearly 50% from 2023 — the market is effectively frozen, and Epic owns the installed base.

Now the AI scribe arrives natively, pre-integrated, and critically bundled into existing Epic licenses. That removes the two biggest adoption barriers for ambient documentation: budget approval and integration engineering. A health system CIO evaluating a standalone scribe startup now has to justify paying for something Epic just bundled into the contract. The startup's integration work is already done inside Epic's walls. Distribution is instant: 325 million patients, 3,300 hospitals, zero sales cycles.

The lawsuits tell the same story. Particle Health alleged in September 2024 that Epic used its EHR market power to "snuff out" competition in adjacent markets. CureIS Healthcare followed in May 2025 with claims of a "multi-prong scheme to destroy" its business, including blocking data access and raising security objections. Epic calls the suits meritless. Oracle EVP Ken Glueck has called Faulkner "the single biggest obstacle to EHR interoperability." But the competitive effect is clear: Epic controls the data layer, the workflow layer, and now the AI layer — all inside one proprietary stack.

For clinical AI startups, the moat just got deeper because the baseline just moved. "Good enough" ambient documentation is now a default feature, not a premium add-on. Startups must prove defensibility beyond integration, including specialty-specific clinical reasoning, quality measurement, revenue cycle automation, or patient-facing workflows that Epic hasn't prioritized. The bar didn't rise incrementally. It jumped.

Startups at the Crossroads

The analyst consensus has fractured into two camps. One argues Epic's 43.7% hospital market share and native distribution will squeeze smaller scribe startups over time. The other contends ambient scribing is becoming a feature, not a product category, leaving room for vendors that go deeper on specialty coverage, multi-EHR support, or capabilities Epic hasn't built yet. Both readings are grounded in the same data — they just weight the variables differently.

Abridge sits at the center of the tension. The company rode its 2023 Epic partnership to a $5.3 billion valuation (Precedence Research reported), $300 million Series E, and over 150 health system customers by mid-2025. Its contracted annual recurring revenue hit $117 million in Q1 2025, The Information reported. Then Epic announced AI Charting, rumored at a price materially below Abridge's range, and quietly sold its Abridge stake. The message was unmistakable: the distribution partner is now the competitor.

Abridge's response has been threefold. First, it broadened EHR support to Oracle Cerner, Meditech, and Athenahealth — a direct hedge against Epic concentration. "They can no longer afford to milk the Epic cow for growth," one observer noted. Second, it expanded beyond scribing into medical coding, real-time prior authorization, and clinical decision support, building its own large language models to power them. Third, it signaled acquisition appetite to accelerate the roadmap. CEO Shiv Rao told Business Insider in August the startup is considering buys to further accelerate growth.

Other startups are picking different lanes. Heidi positions itself as a clinician-first AI partner, not an EHR feature. Its leadership argues documentation is only one slice of a broader workflow layer, and that many Epic customers run Heidi alongside native tools because the integration depth addresses needs Epic doesn't prioritize. Suki and DeepScribe sit at higher price points, betting on specialty-specific workflows and enterprise-grade change management to justify the premium.

The differentiation isn't simply whether a note can be generated — it's whether clinicians actually adopt the system, trust its outputs and use the data beyond documentation to deliver value across the enterprise.

Pricing pressure is already reshaping contracts. Federally qualified health centers negotiating 2026–2027 renewals are inserting Epic-Art-shipping provisions: early-termination triggers if Epic's tool reaches comparable functionality in their OCHIN tenant, pricing step-downs if Epic bundles AI Charting into the base license, and multilingual quality SLAs that recognize Epic's English-only launch leaves a gap for Spanish-speaking panels. OCHIN tenants typically receive Epic features 6–18 months after general availability, creating a negotiated window for incumbents.

Specialty divergence may become the durable moat. Complex fields (oncology, cardiology, surgical subspecialties) operate in highly nuanced, variable workflows where generic ambient capture falls short. Abridge's 28-language support and specialty-tuned models target this directly. Heidi emphasizes cross-workflow support that spans pre-charting, orders, and follow-up. The startups that survive will likely be those that own a clinical problem end-to-end, not those that transcribe it better.

The antitrust overhang adds unpredictability. Epic faces lawsuits from Particle Health and CureIS alleging competitive smothering. If discovery surfaces evidence that AI Charting's rollout was timed or structured to disadvantage partners, the regulatory risk could force behavioral concessions — or at minimum, slow Epic's ability to bundle aggressively.

For now, health systems are dual-running: 90-day pilots comparing Epic Art note quality, Spanish coverage, and provider satisfaction against incumbent vendors. The data from those bake-offs will decide whether scribing stays a contested category or collapses into the EHR baseline.

The Hiring Shock

Epic's native AI scribe launch arrives as the clinical AI labor market is already in violent transition. LinkedIn's 2025 Jobs on the Rise list ranked Artificial Intelligence Engineer first in the United States, with Artificial Intelligence Consultant second — both outpacing every clinical and technical role tracked. The platform examined millions of job transitions from January 2022 through July 2024 to calculate those growth rates. As of November 2024, the University of Maryland and LinkUp's AI job tracker counted 16,591 new AI postings, a 59 percent increase since January 2024, with California, Washington, and Texas absorbing the largest shares. California alone commands roughly 33 percent of all AI engineer postings nationally.

ManpowerGroup reported a 16 percent jump in AI-related postings over a three-month span in 2025 even as overall tech hiring fell 27 percent year-over-year. A separate recruiter survey logged a 59 percent surge in AI job postings for 2024. Eighty-nine percent of companies said AI was creating new roles (MLOps engineer, AI architect, generative AI engineer) while simultaneously worrying about retention and ROI. In healthcare specifically, staffing surveys flag surging demand for data analysis in patient care and imaging, the exact workflows Epic's scribe targets.

Salaries reflect the pressure. Glassdoor's U.S. aggregate puts median total pay for AI engineers around a median with a typical base range. Top-paying companies (Meta, Apple, and their peers) report ranges up to a high figure. A 903-posting analysis from 365 Data Science found an average AI engineer salary, up more than $50K from the prior year. The most frequently cited band was a range (34 percent of postings), followed by another range (18 percent) and above a threshold (13 percent). Experience compounds fast: 0–1 year at a level, 2–3 years at a level, 4–6 years at a level, 7–9 years at a level, 10+ years at a level. Only 2.5 percent of postings target candidates with 0–2 years of experience; 75 percent explicitly seek domain experts over generalists.

First-party board data from Zero G Talent confirms the premium at the frontier. Anthropic added 45 roles in the past seven days with a salary band (median) across 545 salaried positions. Databricks added 33 roles in the same window at a band (median) across 482 salaried positions. Both companies list multiple roles above a threshold for staff and principal engineers — compensation levels that health systems and Epic itself must now contend with when recruiting clinical AI talent.

The skill stack is hardening around a recognizable core. Python appears in 71 percent of AI engineer postings; Java holds at 22 percent; SQL at 17 percent. Cloud deployment is no longer optional: AWS leads at 32.9 percent, Azure at 26 percent, and Kubernetes at 17.6 percent. Model-side, PyTorch (37.7 percent) and TensorFlow (32.9 percent) dominate. But the clinical layer adds distinct requirements. Natural language processing shows up in 19.7 percent of postings (central to scribe architectures), while retrieval-augmented generation (RAG) appears in 13.6 percent, fine-tuning in 14.8 percent, and AI agents in 10.6 percent. LangChain (10.7 percent) and LlamaIndex (4.3 percent) signal the move toward compound systems that chain LLMs with structured clinical knowledge bases. Vector databases register at 4.5 percent, reflecting the same need. Speech recognition, at just 2.0 percent overall, is underweighted in general AI postings but disproportionately critical for ambient scribe teams.

Education requirements are shifting. PhDs appear in 27.7 percent of postings but 48.6 percent accept master's or bachelor's degrees, emphasizing practical experience. Engineering degrees lead at nearly 70 percent of postings; computer science follows at 61 percent. A quarter of listings have no specific degree requirement. The Bureau of Labor Statistics projects 23 percent growth for computer and information research scientists (the category that includes AI engineers) from 2023 to 2033, far above the occupational average.

Remote work remains scarce. Only 5.9 percent of AI engineer postings offer fully remote positions, a constraint that concentrates hiring in the same coastal hubs where Epic's largest health-system customers operate. That geographic lock-in matters: Seattle logged 1,472 AI-related postings as of January 2025, ranking second nationally behind the Bay Area. New York, Boston, Austin, and Raleigh form the next tier.

The Epic launch reshapes demand in two directions. Health systems adopting the native scribe will need integration engineers who understand Epic's backend systems, FHIR APIs, and the Nuance DAX or Abridge middleware — a niche skill set that barely existed two years ago. Simultaneously, the startups displaced by Epic's moat will shed or pivot talent, flooding the market with clinicians-turned-product-managers, prompt engineers tuned to specialty workflows, and ML engineers who have only ever built on top of Epic's sandbox. The net effect is a barbell: deep platform engineers at the top, clinical implementation specialists at the bottom, and a thinning middle.

New roles are crystallizing. Ambient clinical AI engineer, FHIR/ML platform engineer, clinical AI product manager, AI governance and risk officer, and Chief Medical AI Officer appear in postings from health systems and vendors alike. The Stanford-Harvard State of Clinical AI report (January 2026) noted that publication counts for rigorous clinical AI trials rose from near zero in 2015 to 372 in 2025 — evidence that validation and regulatory skill sets are becoming hiring criteria, not just research artifacts. Meanwhile, medical transcriptionist employment is projected to decline 4.7 percent from 2023 to 2033, a direct substitution effect that Epic's tool amplifies.

For job seekers, the message is blunt: domain fluency now outweighs model novelty. A candidate who can ship a RAG pipeline over Epic's FHIR endpoints using LangChain and a vector database, with documented HIPAA compliance and clinician-in-the-loop evaluation metrics, commands a premium over a generalist LLM fine-tuner. The market is paying for deployment maturity in a regulated workflow — exactly the moat Epic just deepened. But it tightens: specialized roles in validation, safety monitoring, and EHR-integrated model ops will command premiums over general ML engineering.

Clinicians' Choice

Adoption is real, but it's lumpy. A Nature Medicine survey of 598 UK GPs found 40 percent report current AI scribe use and another 23 percent have tried one; usage ranged from 5 to 100 percent of consultations with a mean of 60 percent. In the US, the Elsevier Clinician of the Future survey found 48 percent of clinicians using AI for work, up from 26 percent a year earlier, but 97 percent of those users lean on generalist tools like ChatGPT while only 76 percent have touched a clinical-specific product. The gap matters: clinicians trust what they've tested, and most have tested the wrong thing.

The Permanente Medical Group's 7,260 physicians across 2.5 million encounters offer the clearest signal yet. High-frequency users (the top third by activation count) captured 89 percent of all scribe sessions and saved two and a half times more documentation time per note than low-frequency peers. Physicians reported statistically significant drops in after-hours "pajama time," total note-taking minutes, and time per appointment. Forty-seven percent of patients said their doctor spent less time staring at the screen; 39 percent said they got more direct conversation. Eighty-four percent of physicians rated the impact on patient interactions positive, 82 percent on work satisfaction. The tool works, for the doctors who use it heavily.

But the median clinician isn't a high-frequency user. Adoption skews male (odds ratio 1.64), toward private practice (OR 2.88), toward GPs running five to six sessions weekly (OR 2.17), and toward trainers (OR 3.10). The most common reason to pick a scribe? A colleague's recommendation. The most common reasons to quit? Lack of practice support and medicolegal liability fears. Over 60 percent of surveyed GPs, including current users, agree inaccuracies, errors, and misinterpretations create medicolegal threats. Over half flag workflow disruption and poor EHR integration as risks. Only 40 percent of past and never-users believe scribes reassure patients.

The integration question is where Epic's native play bites. NHS England issued a cease letter in June 2025 telling GPs not to use scribes unless they meet NHS data-management and patient-safety standards, explicitly placing liability on the individual user or practice. Professional bodies now advise using only products registered as medical devices with completed clinical safety and data-protection assessments. Yet reports persist of unregulated scribes in live use with inadequate patient consent. The regulatory vacuum is real: no national guidance, no standardized evaluation, no coordinated implementation framework.

Clinicians want trust signals. Sixty-eight percent say auto-cited references would increase confidence; 65 percent want training on peer-reviewed content; 64 percent demand current resources. In the US and UK, 75–81 percent rank factual accuracy as the top requirement. Yet only 32 percent feel their institution provides adequate AI access, 30 percent report sufficient training, and a mere 29 percent say governance structures exist. The demand is there; the scaffolding isn't.

Vision-enabled scribes (processing video, not just audio) hit 98 percent overall accuracy versus 81 percent for audio-only, primarily by cutting omissions (10 errors vs 358). But they introduce new privacy dynamics: continuous video recording in exam rooms, differential toxicity scores across racial labels in simulation studies, and unresolved consent workflows. The CHAI governance framework and ADS evaluation tools are emerging, but prospective studies with oversight remain scarce.

Epic's February 2026 general availability of native AI Charting changes the procurement calculus. Health systems are already trimming vendor rosters as finance leaders weigh ROI against Epic's expanding bundle. For the clinician, the choice collapses to a simpler question: does the tool embedded in the EHR I already live in solve the pajama-time problem without creating a liability problem? The answer, for now, depends on which health system pays for the license and whether the implementation team bothers to train the physicians.

What Comes Next

The regulatory calendar is the first place to watch. FDA's 2026 guidance agenda, published February 2026, explicitly lists two draft guidances slated for this year: "Use of Digital Health Technologies in Clinical Investigations of Drugs and Biologics" and "AI/ML Quality Considerations in Pharmaceutical Manufacturing" — both categorized as "New (Draft) Planned 2026." CDER's FRAME initiative, which held a joint workshop with PQRI in September 2023 and synthesized public comments in an AAPS Open summary (May 2025), is the engine behind the manufacturing guidance. Stakeholders emphasized "good data management, model validation best practices, and uncertainty about integrating AI into existing Quality Systems," along with concerns about "explainability of complex models, use of third-party data, and meeting CGMP requirements with novel technologies." The consensus: "AI models need robust lifecycle plans and should align with current strategies for drug quality." Given FDA's comment periods, the manufacturing guidance could land as a Draft Level 1 by end of 2026 or a Final Level 1 if discussion-paper feedback suffices. In parallel, CDRH's AI/ML device guidance on change control (planned for mid-2025) will set industry-wide expectations for AI lifecycle management that could influence drug-device combination products like Epic's Nuance/Abridge integration.

Commissioner Makary's January 2026 remarks frame this year as a "pivot point": cutting "27 guidances … by 50%" and streamlining rules on software and AI. Industry associations (PhRMA, BIO) have issued white papers urging swift finalization, noting "companies avoid investing in large-scale AI systems without regulatory clarity." The Digital Therapeutics Alliance is lobbying for expedited pathways for drug–software combinations, similar to orphan drug vouchers. Patient advocates and ethicists counter that "under-tightening policies too much could risk data privacy or equity in AI models." FDA appears to be balancing these views through stakeholder workshops and public comments, including the March 31, 2026 RFI on the aforementioned topic that explicitly seeks input on "verification/validation of novel sensors, digitally derived endpoints in specific disease areas, and ideas for future public workshops." The TEMPO pilot (Technology-Enabled Meaningful Patient Outcomes), announced December 2025, lets manufacturers request enforcement discretion for premarket authorization on certain digital devices for chronic diseases — a template that could expand if successful.

On the clinical front, the next 12–18 months will test whether specialized clinical AI tools can maintain their safety edge over general-purpose LLMs. The NOHARM benchmark (Stanford/Harvard) found severe-harm potential up to 24.6% for 20 generalist LLMs tested against 1,100 real case-based tasks, with over 80% of severe errors being errors of omission. By contrast, specialized tools (AMBOSS LiSA, Doximity Ask, OpenEvidence, Glass Health) scored severe-harm rates of 2.9% to 5.4%. Mount Sinai found ChatGPT Health under-triaged 52% of genuine emergencies; a BMJ Open audit of five major chatbots found nearly half of answers to common health questions contained misleading or problematic information. Meanwhile, only about 5% of AI-enabled medical devices had reported adverse-event data by mid-2025, including at least one death linked to device malfunction. These numbers will pressure health systems to demand rigorous validation, not just explainability, before deploying scribes at scale. The Santa Clara ethics analysis argues regulations must "first prioritize rigorous validation, then inherent interpretability, and finally post-hoc explainability only for global model audits," warning that the January 2026 FDA exemption of certain clinical decision support tools from regulation "represents an ethical crisis as clinical AI continues to be integrated into the US health care system."

Adoption curves suggest the market is splitting. Generative AI adoption among US healthcare organizations rose from 25% in late 2023 to 47% in 2024, reaching 50% by end of 2025, with over 80% of surveyed leaders reporting at least one deployed use case. But a national survey of 2,174 nonfederal acute care hospitals found 31.5% already using generative AI integrated with their EHR by 2024, 24.7% planning to adopt within a year, and 43.7% delayed or uncertain. Adoption correlates strongly with resources: major teaching hospitals (53.9%) and system-affiliated hospitals (38.5%) were far more likely early adopters than independent hospitals (16.3%). Epic's 85% penetration across its Art, Emmie, and Penny copilot tools means the installed base for native scribe rollout is massive — and health systems may delay third-party vendor selections pending Epic's full feature set. That dynamic forces AI scribe startups to prove defensibility beyond EHR integration: specialty-specific workflows, measurable ROI on clinician time, or safety certifications that Epic's general-purpose copilots lack.

Talent demand will track these splits. Eighty-one percent of US physicians report using AI professionally in 2026 (up from 38% in 2023), averaging 2.3 distinct use cases each (up from 1.1). Yet only 16% use AI for actual clinical decision-making — a gap between broad experimentation and deep integration. Eighty-eight percent worry about skill loss from over-reliance; 85% want a direct say in adoption. The AAMC projects a shortage of up to 86,000 physicians by 2036, driven by population aging (Americans 65+ expected to outnumber children under 18 by 2034). AI that demonstrably reduces documentation burden (ambient scribes cutting after-hours charting) becomes a retention lever, not just a productivity tool.

Two late-stage clinical readouts will signal whether AI-discovered drugs can break the Phase 2 ceiling. Insilico Medicine registered a 320-patient Phase III study (NCT07687459) for rentosertib (TNIK inhibitor for IPF) on July 7, 2026, with estimated start August 30, 2026. Generate:Biomedicines' AI-engineered antibody GB-0895 started Phase 3 SOLAIRIA-1 on December 3, 2025. On July 29, 2026, FDA granted Insilico's ISM6331 (pan-TEAD inhibitor for advanced mesothelioma) Fast Track Designation. The 2024 BCG analysis found AI-discovered molecules had 80–90% Phase 1 success but only ~40% in Phase 2, "comparable to historic industry averages." The 2025 Nature Medicine rentosertib paper cautioned that "AI-discovered drugs have experienced similar levels of phase 2 trial failure as non-AI-discovered drugs" and none had yet progressed through Phase 3. If rentosertib or GB-0895 clears Phase 3, the narrative shifts from "preclinical acceleration" to "clinical translation" — unlocking the next wave of capital into AI-native biotechs and, by extension, the clinical AI infrastructure they require.

The global AI-in-healthcare market was valued at a figure in 2025 and is projected to reach a figure in 2026, growing to a figure by 2033 (38.9% CAGR). North America held a share of revenue in 2025. McKinsey estimates global spending on AI in life sciences (R&D and manufacturing) at a figure annually, projected to double by 2030. ISPE's 2024 survey found ~20% of member companies conducting AI pilots in manufacturing, up from 5% in 2020. An internal CDER audit (November 2025) found 60% of NDAs in 2024 contained at least one AI-generated analysis; ~10% of IND applications now include trial protocols with remote monitoring plans. Among Phase III drug trials, ~15% incorporated wearable or app-based data as of 2024, a three-fold increase over five years.

Faulkner's staff still tells suitors to leave their cars running. But the CIOs who once asked when the EHR would just do this now have their answer — and the 120-plus startups that colonized the workflow have to decide whether to fight for the specialty niches Epic hasn't reached, or sell to the platform that just made their core feature bundled.

Comparable Figures at a Glance

Working in AI? Zero G Talent tracks the openings: see every open Databricks role, browse AI jobs, openings at Anthropic, and the people building the field.