AI in Healthcare | What Comes Next Ep. 12

“Our thesis is that the cost of writing code is heading towards zero. The cost of producing meaningful software is not. […] The biggest risk with AI isn’t that it’s wrong. It’s that it sounds right.”

— Ashby Engineering Blog, “AI, Ashby Engineering, and the Future” (June 2026)

StatusGo’s read: Payers are the year’s quiet winners. Centene, Humana, and UnitedHealth are all up double digits (median +19.7%), while acute-care operators HCA, Tenet, and Universal Health each shed 18% or more. And AI infrastructure is no monolith: AMD’s +118% masks double-digit losses at Microsoft, Meta, and Tesla.

Mayo Clinic and Microsoft are co-developing a frontier AI model purpose-built for healthcare, trained on Mayo’s de-identified clinical data and longitudinal insights. Notably, Mayo will own the model, while Microsoft distributes it via Azure Foundry APIs. The move signals provider-owned, clinically validated models challenging general-purpose AI. Mayo Clinic News Network

Dr. Michael Jerkins argues AI exposes, not causes, a drift in medical training: models now ace the USMLE, yet accuracy drops sharply with real patients. With medical literature doubling every 73 days and curricula lagging, he urges schools to teach AI judgment and select for human skills. MedCity News

Adaptive Innovations raised $60 million, including a $50M Series A led by Felicis, to run as a full-stack, AI-native home health provider, not a software vendor. It reports 100,000+ visits and 80% less clinician documentation time. For health systems, it is a discharge partner that accepts patients legacy agencies reject. Pulse 2.0

A STAT essay argues healthcare’s biggest AI deployment is billing, not diagnosis: hospital coding AI and insurer denial AI battle over every record. BCBS found AI-driven upcoding could add $2.3 billion nationally, while health systems spend over $140 billion yearly on revenue cycle work. The friction lands on patients. STAT News

Stellarus, spun out of Blue Shield of California, launched AtlasIQ, a platform unifying clinical, claims, and operational data into a real-time, AI-ready source of truth. It already handles 530 billion records for the insurer. The pitch targets the data-infrastructure bottleneck that stalls payer AI at scale. PR Newswire

Over 1 million children have lost insurance since 2025, and roughly another million may follow as new Medicaid rules and immigration fears bite, leaving nearly 2 million uninsured, per Georgetown. The One Big Beautiful Bill Act is driving the erosion. For payers, it means shrinking Medicaid and ACA membership. KFF Health News

Anthropic, maker of Claude, confidentially filed a draft Form S-1 with the SEC on June 1, setting up a possible IPO pending review. The move follows its $965 billion Series H valuation. For health systems betting on frontier AI, a public Anthropic means more financial transparency and vendor scrutiny. Anthropic

President Trump signed a narrowed AI order asking firms to voluntarily submit frontier models for federal review 30 days before release, down from a 90-day draft scrapped after industry pushback. A companion measure would route AI-found software vulnerabilities to critical infrastructure operators, hospitals included. Politico

Over 220 former unicorns dropped below billion-dollar valuations as AI funding crowds out everyone else, per PitchBook. Companies that last raised in 2021 are worth 68% less on average, with enterprise SaaS hit hardest. Healthtech vendors built before generative AI face the same fundraising squeeze and survival risk. CNBC

A Tsinghua study of 953 economics papers found AI-written work trails humans mostly on idea quality, driving about 71% of the gap, while execution lags far less. Only 0.8% of AI papers beat the median human on both. For healthcare AI, the ceiling is original thinking, not technical execution. SSRN

Huawei chairman Xu Zhijun thanked Washington for export controls, saying sanctions forced China to build its own chip stack, then unveiled a 1.4nm-class roadmap targeting 2031. A credible AI-accelerator supply chain outside US jurisdiction reshapes long-term compute cost and availability for any health system scaling AI. Tom's Hardware