AI in healthcare risks repeating the mistakes that made EHRs a burden

After leaving her role as interim CEO of Veradigm, a Chicago-based health IT data analytics company, S. Yin Ho began writing a book tracing the last quarter-century of health IT. Over 18 months of research, she came to the conclusion that EHRs, the infrastructure meant to modernize clinical care by documenting and communicating a patient’s needs, were hijacked instead to serve billing, compliance, and revenue cycle operations. 

In “Rushing Headlong: Health IT’s Legacy and the Road to Responsible AI,” published in December 2025, Ho argues that artificial intelligence is now being layered onto that same foundation and risks reproducing the same distortions.

Ho is a Yale-trained emergency physician with an MBA from Harvard Business School who has spent two decades in health IT leadership, including roles at Pfizer, Medidata, and Aetion. At Veradigm, she led the $140 million acquisition of ScienceIO to integrate generative AI with electronic health record (EHR) data spanning more than 200 million patients. Her argument is not that AI lacks clinical promise, but that the economic forces shaping its deployment are pulling it toward the same transactional purpose that defined EHRs.

New data from Digital Health New York’s 2026 Healthcare Innovation Report supports Ho’s concern. In New York, 126 healthcare companies raised $4.8 billion in 2025, a 20 percent increase over the prior year. Administrative efficiency led all funding categories at 25 percent of total capital. 

How EHRs became billing-driven systems

In a Q&A in the DHNY report, Ho explains that when EHRs were co-opted for billing purposes, clinicians began documenting to justify reimbursement rather than to inform one another and support sound clinical reasoning, producing two parallel systems with little meaningful connection.

A 2024 study in the Annals of Family Medicine found that primary care physicians’ EHR workload has continued to grow, with total EHR time per eight hours of scheduled clinic time rising nearly 8 percent between 2019 and 2023. Time spent on orders rose 58.9 percent and inbox management rose 24.4 percent over the same period, while after-hours EHR work also increased significantly. EHR design has broadly prioritized billing and administrative functions over clinical decision-making, a pattern Ho argues is now being replicated in AI.

Ho advocates building smaller, domain-specific AI models trained on curated clinical data with governance built-in from the start. She warns that if AI is deployed mainly for billing, prior authorization, or compliance, the result will be greater administrative complexity and further drift from patient-centered care.

Healthcare AI is following the money

The ambient AI scribe market shows how these incentives play out. These tools use generative AI to transcribe clinical encounters and draft notes. By mid-2025, 62.6 percent of Epic hospitals had deployed ambient documentation and the category generated roughly $600 million in revenue, more than double the prior year.

Ambient scribes can reduce after-hours charting, lower burnout, and improve patient interaction. But vendors are increasingly marketing these products as revenue cycle tools. Ambience Healthcare’s mid-2025 funding round positioned its platform as a driver of coding integrity and revenue-cycle performance. A policy brief in npj Digital Medicine warned that if revenue optimization becomes ambient AI’s primary use case, the industry could trigger a “coding arms race” in which AI-assisted billing practices grow more aggressive and complex, adding administrative friction without improving patient care.

A January 2026 JAMA Health Forum analysis found that ambient scribes were championed as a fix for documentation burden, itself an unforeseen consequence of earlier health IT deployment. The authors noted that AI-driven documentation completeness could trigger payer downcoding responses. Cigna began automatically reducing certain higher-level claims in October 2025 unless documentation clearly supported the complexity.

What the AI market shift means for health systems

The DHNY report points to 2026 as a year of consolidation. Its annual Digital Health 100 list turned over nearly half its roster, adding 48 new companies. DHNY CEO Bunny Ellerin described the shift as one from experimentation to execution, and several contributors echoed the expectation that health systems would demand measurable ROI from digital health partners.

Ho’s counterpoint is that defining ROI too narrowly becomes the problem. In her book, she argues that EHR certification incentives in the late 2000s created a gold-rush dynamic that concentrated market power among a few vendors and locked clinicians into systems that did not serve them.

A similar consolidation is visible in AI. According to Rock Health, average deal size for U.S. digital health startups rose to $29.3 million in 2025 while total deal count fell, and mega-rounds over $100 million accounted for 42 percent of all investment.

Ho does not reject AI in healthcare, she sees generative AI as a genuine opportunity to bridge clinical care and research. Her argument is about trajectory: the same incentive structures that turned EHRs into billing infrastructure are now shaping which AI tools get funded and embedded in clinical workflows.

Across the DHNY report’s 30 contributor perspectives, administrative efficiency and revenue cycle automation appear far more often than clinical decision support or research integration, raising a broader question for health system leaders: not just what a tool automates, but what priorities it reinforces.