New AI OS in Health: Abridge, OpenEvidence, Heidi, and Epic are racing to build the AI operating system

The new AI operating system in healthcare combines three pillars: AI Scribe (ambient clinical documentation), AI Billing (automated coding and revenue cycle management), and AI Clinical Decision Support (evidence-based recommendations at the point of care). These are linked through an integrated workflow layer that sits atop legacy EHR/EMR systems.

This intelligence layer transforms the physician's interaction model: voice-input replaces manual data entry, and automated downstream workflows handle billing, coding, and clinical recommendations. The result is an AI-native operating system that does not replace the EHR but makes it invisible to the clinician.

Each company started from a different entry point. Abridge (founded 2018, $5.3B valuation, 770M funding) began with the AI Scribe as its core product through Epic integration, later added billing, and in April 2026 partnered with NEJM and JAMA to add clinical decision support.

OpenEvidence (founded 2022, $12B valuation, 40% of US physicians) started with AI-CDS, building the first Med-GPT at scale based on NEJM and JAMA content. It launched its own scribe called Visits in August 2025, and partnered with Tandem for payer billing in April 2026.

Heidi (founded 2019 in Australia, $465M valuation, 116 countries) started with the AI Scribe supporting 110 languages, launched Heidi Evidence for clinical decision support in February 2026, and partnered with R1 RCM for US market billing.

Epic (founded 1979, #1 EHR globally, 700 health systems, 50% of US hospital beds) has billing native since inception, launched its own AI scribe with Microsoft in August 2025, and launched Curiosity for AI-CDS in September 2025.

Abridge's partnerships with the New England Journal of Medicine and JAMA Network are significant because these are the same journals backing OpenEvidence. The critical question is who holds the real power in this ecosystem: the LLMs like OpenEvidence covering 45% of US physicians, the medical journals publishing the content, or the incumbents like Epic controlling distribution.

These content deals are not exclusive, which creates vulnerability. If NEJM and JAMA license to everyone, the competitive moat shifts from content access to workflow integration and clinical adoption. The company that owns the physician's daily workflow, not the one with the best content license, will likely capture the most value.

Epic cut off Abridge's preferred integration path and launched its own competing scribe with Microsoft. This move demonstrates that Epic will never be a true ecosystem integrator. In a technology ecosystem, trust is the new currency. Breaking that trust has consequences: it pushes health systems to evaluate whether their strategic dependency on Epic is sustainable.

Meanwhile, in Europe, Nexus AG (Germany's #4 hospital information system) took the opposite approach: acquiring Omilon to create a data layer that orchestrates third-party AI solutions rather than building competing products. This open-platform model may prove more sustainable than Epic's walled-garden strategy.

The AI OS market will consolidate. This stack is becoming a critical infrastructure piece for both providers and payers, and the OpenEvidence x Tandem partnership shows how the AI layer is starting to link provider and payer workflows.

Private equity continues to invest in legacy systems, betting that the transition will be slow. But the pace of convergence (all four players moved into all three pillars within 18 months) suggests the window for legacy systems is narrowing faster than expected. The winners will be the companies that can deliver all three pillars seamlessly while maintaining trust with health systems.