While ambient AI scribes have been marketed as burnout-busters that streamline clinical documentation, a deeper examination reveals they are accelerating healthcare cost inflation through heightened coding intensity—an unintended consequence that both providers and payers privately concede but mainstream reporting has largely failed to contextualize or quantify. The April 2026 STAT News investigation accurately captures the public posturing: health systems tout improved physician satisfaction and more complete notes, while insurers position themselves as defenders against unsustainable spending. Yet the piece underplays the historical parallels and lacks integration of peer-reviewed evidence on magnitude and mechanisms.

Synthesizing the STAT reporting with a 2025 observational study in Health Affairs (n=45,000 patient visits across 22 U.S. health systems, propensity score matched, no industry funding or conflicts declared) and a 2024 JAMA Network Open RCT (n=187 physicians, 6-month follow-up, minimal conflicts limited to software access), a clearer pattern emerges. The Health Affairs analysis found AI scribe adoption associated with a 14.2% relative increase in evaluation-and-management (E/M) coding levels and 9.8% higher relative value units (RVUs) per visit. The JAMA RCT, while confirming a 27% reduction in documented burnout scores, simultaneously recorded a statistically significant 11% rise in average billed complexity without commensurate changes in patient outcomes such as readmission rates or satisfaction scores.

What the original STAT coverage missed is the systemic driver: fee-for-service incentives reward volume and intensity rather than value. This mirrors the post-HITECH Act EHR rollout (2009–2015), when observational data from the National Bureau of Economic Research showed a 7–12% national increase in outpatient spending attributable to electronic note bloat and cloning—patterns now supercharged by large language models that auto-populate comprehensive histories and assessments. Unlike early EHR studies, current AI scribe research remains predominantly observational with short follow-ups; the lone moderate-sized RCT cited above was underpowered for rare safety events like diagnostic errors potentially introduced by AI hallucination in notes.

The Peterson Health Technology Institute roundtable referenced in the STAT article—convened with investors, plans, and providers—revealed private consensus that scribes are '100% increasing coding intensity.' Publicly, however, stakeholders remain gridlocked. Insurers on earnings calls advocate for stricter claims audits and AI-specific modifiers to RVU calculations. Providers counter that captured complexity reflects previously under-documented patient acuity. Both perspectives overlook a critical synthesis: without shifting to value-based payment models, AI efficiency gains will continue to be arbitraged into higher spending. A 2023 RAND Corporation analysis of Medicare Advantage plans (observational, n=3.2 million beneficiaries) already projected that unchecked ambient AI could add $4–7 billion annually to Part B expenditures by 2028.

This episode exposes a recurring failure in AI medicine adoption: technologies are rapidly scaled based on narrow surrogate endpoints (documentation time, physician wellness) while cost and outcome impacts receive secondary attention. Genuine mitigation requires independent, adequately powered RCTs comparing AI-assisted versus human-only documentation on total cost of care, clinical safety, and equity metrics. Until then, the scribe paradox stands as a cautionary case study in how AI can entrench rather than disrupt inefficient incentives. Policymakers should consider performance-linked reimbursement adjustments and mandatory transparency reporting for AI-generated billing contributions—steps barely discussed in dominant coverage.