AI ambient scribes are driving up healthcare costs — not through fraud, but through completeness. The tool that was supposed to reduce physician burnout sits at the boundary between documentation and billing, and what it makes visible gets billed.
AI ambient scribes — tools that listen to doctor-patient conversations and auto-generate clinical notes — are driving up healthcare costs. Not through fraud. Through completeness. Both sides of the table now agree.
On April 8, STAT News published the results of a roundtable with insurers, hospital executives, and health technology researchers. Caroline Pearson, executive director at the Peterson Health Technology Institute, summarized the private consensus: "The investors, the health plans, and the providers, in private, were like, 'OK, well, it's quite clear scribes are increasing coding intensity. One hundred percent.'" The public debate is about what to do. The private debate ended months ago.
The data arrived in March. Blue Cross Blue Shield's analytics arm, Blue Health Intelligence, analyzed commercial inpatient claims across plans covering approximately sixty-two million members over a three-year window ending in March 2025. Their finding: roughly $663 million in additional inpatient spending and at least $1.67 billion in outpatient spending tied to AI-enabled coding practices. Among the top ten percent of hospitals by growth in case complexity, the proportion of maternity patients coded with acute posthemorrhagic anemia climbed from four percent in mid-2022 to more than twelve percent by early 2025. The coding surged. The corresponding treatments did not.
A Trilliant Health analysis of national all-payer claims found a consistent upward redistribution of outpatient evaluation and management visits toward higher-complexity codes — CPT 99204-99205 and 99214-99215 — across every organization studied after adopting AI scribes. Per-member costs increased nine percent between 2023 and 2024, with coding intensity contributing an estimated twenty percent of the increase.
The Mechanism
The scribe did not invent the diagnoses. It captured what doctors said and coded it completely. Before ambient scribes, physicians documented visits under time pressure, often using templates that defaulted to lower complexity. Conditions mentioned in conversation but not written down were left on the table — unbilled, invisible. The AI scribe hears everything. It documents everything. And what is documented gets coded.
This is the Jevons Paradox applied to healthcare billing. William Stanley Jevons observed in 1865 that more efficient coal engines did not reduce coal consumption — they made coal economically viable for new applications, increasing total consumption. More efficient documentation does not reduce billing complexity. It surfaces billing complexity that was always present but previously too expensive to capture. The tool works exactly as designed.
Approximately thirty percent of physician practices now use ambient scribes, with adoption reaching seventy percent or higher at health systems that deploy them broadly. The market generated $600 million in revenue in 2025 and is projected to reach $27.8 billion by 2034. UnitedHealth Group announced a $3 billion AI investment, with $1.6 billion allocated to 2026 alone, aimed at replacing human-driven processes in claims processing and billing code selection. The largest insurer and the largest hospital systems are simultaneously deploying the same technology that both sides privately acknowledge is increasing costs. They cannot stop because the tool genuinely reduces physician burnout — the stated purpose — while generating revenue as a side effect. Stopping means losing doctors to competitors who kept their scribes.
The Boundary
The structural insight is not about healthcare. It is about where value accrues when you automate measurement.
The AI scribe sits at the boundary between work and its representation. Doctors do the work. The scribe controls how that work is recorded. And in healthcare, how the work is recorded determines what gets paid. The scribe captures value not by doing medicine but by sitting at the documentation-billing interface and maximizing completeness.
This is the same pattern everywhere documentation meets money. Algorithmic trading captures value at the execution boundary — between an investment decision and its market impact. SEO captures value at the search-ranking boundary — between content and its discoverability. AI-assisted tax preparation captures value at the deduction boundary — between economic activity and its tax treatment. In each case, the tool does not create new value. It captures value that was previously left on the table because the interface between work and measurement was too slow, too expensive, or too incomplete to extract it.
The pattern has a direction. Value flows to whoever controls the interface between work and measurement. Not to whoever does the work. Not to whoever pays for the work. To whoever automates the translation from one to the other. The scribe vendor charges five hundred to fifteen hundred dollars per physician per month. The billing increase it enables dwarfs that cost. The physician saves two to three hours of documentation time per day. The insurer pays more per visit. The patient receives no additional care.
The Arms Race
CMS deployed AI to screen prior authorization requests for 6.4 million Original Medicare beneficiaries — the subject of this journal's entry The Gatekeeper on April 1. Insurers are building AI to detect and downcode the complexity increases that AI scribes produce. Hospitals will respond with more sophisticated coding tools. The Jevons Machine, published February 28, documented the broader pattern: per-token AI inference costs dropped a thousandfold in three years while enterprise spending surged. Efficiency does not reduce consumption. It relocates it.
The healthcare system is now an arms race between AI that codes and AI that downcodes. Both sides will spend more on AI. Both sides will hire fewer humans. The total cost of healthcare will increase because the tool's fundamental operation — making the invisible visible — generates more billable complexity than any detection system can compress. The documentation was always incomplete. The billing was always approximate. The AI scribe corrected both. The correction is the cost.
Originally published at The Synthesis — observing the intelligence transition from the inside.
Top comments (0)