AI in Healthcare | What Comes Next Ep. 7

“AI will touch everything everyone … everywhere. And that is not your burden. It’s your opportunity.”

— Alan McKim, Billionaire Founder of Clean Harbors Environmental Services

Backed by Thiel, Altman, Armstrong, and Andreessen, a wave of startups is pursuing embryo screening and gene editing to optimize children for intelligence, framed as a hedge against superintelligent AI. Companies like Nucleus and Preventive claim disease-prevention goals, but scientists warn polygenic IQ gains are minimal and germline editing risks are unknown. For health system leaders, the capital flows and regulatory vacuum signal a coming collision between reproductive medicine, AI ethics, and equity. Mother Jones

a16z argues that current LLMs are frozen at deployment, retrieving from external memory rather than truly learning. Continual learning, where models compress new experience directly into their parameters after deployment, is the next frontier. For health system leaders deploying AI at scale, a model that retrieves your prior authorization history is fundamentally different from one that internalizes your clinical patterns over time. Andreessen Horowitz

A real-time tracker ranking 37 AI models by quality-adjusted tokens per dollar finds ChatGPT Plus at $20 per month delivers the equivalent of $547 per month in API value for heavy users, a 27x multiplier. Claude Max 20x at $200 per month reaches $3,140 per month in API-equivalent value. For health system leaders making AI vendor and licensing decisions, the gap between subscription and API economics is now large enough to be a budget line item. Desktop Commander

A Peterson Health Technology Institute report finds AI is accelerating transaction volume without fixing structural inefficiencies, potentially raising costs. More prior auth submissions, more billing activity, and AI-assisted coding driving higher reimbursement are all contributing to medical cost inflation. Healthcare’s $350 billion annual administrative waste problem remains largely untouched. For CFOs expecting hard ROI, the current deployment pattern is a warning sign. Healthcare Finance News

Cigna is exiting its ACA individual exchange business at year-end, covering 369,000 lives, following its 2025 sale of Medicare Advantage to HCSC. Q1 profit hit $1.65B, up 25% year over year, with Evernorth revenue rising 9% to $58.4B. The medical loss ratio dropped to 79.8%. The moves signal Cigna’s pivot to a pharmacy and benefits services model over traditional insurance, reshaping how it competes with UnitedHealth and CVS. Becker’s Hospital Review

Hospital prices have risen 281% since 2000, three times the inflation rate, driven by government-designed incentives that reward consolidation over competition. CON laws, site-payment differentials, and the ACA ban on physician-owned hospitals protect incumbents, while $1.6 trillion in annual hospital spending flows through a system that subsidizes inefficiency. Site-neutral payment reform is the sharpest lever available. Paragon Health Institute

A Harvard and Stanford study published in Science found OpenAI’s o1 preview matched or exceeded expert physician performance across triage, diagnostic reasoning, and case management in 76 live emergency room cases. AI was strongest at initial triage when information was most limited and excelled on rare and complex cases. Researchers are calling for prospective clinical trials and warn against using these results to cut physicians out of care decisions. Harvard Magazine

At 82, Faulkner still runs Epic, the EHR platform covering 80% of Americans, from a 1,700-acre Wisconsin campus she built without venture capital, acquisitions, or a public offering. Revenue hits $5.8B annually, yet profit is a side effect, not a goal. For health system leaders, Epic’s structure and longevity raise pointed questions about what healthcare IT looks like when mission drives the model. Freakonomics

Tempus AI and Keck Medicine of USC announced a collaboration spanning genomic diagnostics, AI-driven clinical trial matching, and care gap identification across USC’s Southern California network. The partnership targets over 1.5 million annual patient visits and includes joint development of AI tools connecting research to bedside care. For provider leaders, it is a concrete model of how academic medical centers are operationalizing precision medicine at scale. Tempus AI

Mayo Clinic’s AI tool REDMOD correctly identified 73% of hidden pancreatic cancers in CT scans originally read as normal, nearly twice the detection rate without AI, and nearly three times better on scans taken more than two years before diagnosis. The study analyzed close to 2,000 CT scans across multiple hospitals and scanning systems. For provider leaders, this is a concrete signal that AI is moving from administrative efficiency into the hardest clinical problems in oncology. RTT News