In early 2026, somewhere in a busy primary care clinic in Ohio, a physician finished a 22-minute visit with a 67-year-old patient managing diabetes and early-stage heart failure. She said goodbye, clicked a button, and within seconds a complete clinical note appeared in the EHR—inclining the relevant diagnoses, flagging medication interactions, suggesting a prior authorization for a new cardiology referral, and populating the billing code. No typing. No dictation. No pajama time spent charting after hours. That physician's experience is no longer exceptional. It represents a new baseline for what healthcare AI can deliver at scale.
Abridge, the Pittsburgh-based ambient documentation company founded in 2018—before the LLM boom made such timing look prescient—projects it will support over 80 million patient-clinician conversations across 250 US health systems this year. The number lands with weight precisely because it is not a pilot. It is not a demo in a controlled research environment. It is a workflow operating in the complex, messy, high-liability reality of American healthcare, across 28 languages and 50 specialties, in community clinics and academic medical centers alike.
The journey from ambient scribe to clinical intelligence layer is one of expansion along an existing axis. Abridge entered healthcare through documentation—the unglamorous, high-volume problem of clinical note-writing that consumes an estimated 40% of a physician's workday. Listen to the conversation, extract the structure, generate the note, reduce the clerical burden. The wedge was time. The depth came from understanding that documentation is not isolated: it connects to billing, prior authorization, quality metrics, follow-up protocols. Solve the note, and you touch everything downstream.
That understanding of systemic entanglement is what separates Abridge's trajectory from a simple software company adding AI features. When the company began building prior authorization automation—a process that routinely consumes 10-20 hours per physician per week—the infrastructure to do it well was already there. Clinical notes encode the justification. Billing codes encode the procedure. The AI doesn't need to start from scratch; it needs to connect dots that were always meant to be connected but lived in separate systems.
The technical challenge of building a foundation model for clinical language is substantial. Healthcare demands specificity that general-purpose LLMs struggle to deliver: the right diagnosis code, the clinically relevant detail, the note structure that a specialist in pulmonary fibrosis or pediatric cardiology actually needs. Abridge has built specialty-specific evaluation pipelines and employs clinician-scientist teams to test outputs against ground truth. At 80 million conversations, the evals are not academic exercises—they are continuous quality assurance operating at a scale that produces statistically meaningful signal.
The $300 million funding round in June 2025, valuing the company at $5.3 billion, tells part of the story. But the more revealing metric is where that money is going: not into customer acquisition for a single product, but into building the infrastructure for a clinical intelligence layer—AI that acts before, during, and after the patient conversation. The aspiration is a system that anticipates what a clinician needs, surfaces it at the right moment, and disappears when it is not needed. Chai Asawa, Abridge's CTO, has described this as making AI feel like air conditioning: always present, always adjusting, rarely noticed unless something goes wrong.
The question healthcare AI critics raise is legitimate: Can any AI truly understand a clinical conversation? The honest answer is that Abridge is not claiming to replicate physician judgment. It is claiming to offload the cognitive overhead of documentation, coding, and administrative coordination so that physicians can exercise that judgment more freely. At 80 million visits, the data suggests that claim is holding. Clinicians report recovering 10-20 hours per week—time that goes back to patients or, realistically, to the cognitive rest that prevents burnout.
Healthcare has seen waves of technology promises before. EMRs were supposed to reduce overhead. Scribes were supposed to fix charting. The history is littered with solutions that solved one problem and created three more. What Abridge's scale suggests is that the problem worth solving—protecting and enhancing the patient-clinician conversation—may finally have found the right technological form factor. The 80 million visits are not the ceiling. They are the floor.