|
3, 2, 1: Health AI Brief
Every Friday
July 24, 2026
|
|
AI is reshaping healthcare fast. Below are 3 key AI developments, 2 studies, and 1 takeaway for this week to help you better lead with AI. Target read time: 5 minutes. |
|
3
Market Signals
Candid Health closed a $120 million Series D on July 22, led by Sixth Street Growth with Oak HC/FT, 8VC, and Y Combinator participating, at three times its February 2025 valuation. The platform unifies clinical, billing, and provider data, then uses a rules engine plus AI agents to automate work that billing teams do by hand. The company says it processes roughly $7 billion in annual claims for 200+ customers, with contracted run‑rate revenue up 190% year over year. My Take
AI that can show ROI fast is shining right now, i.e., ROI from better billing shows up in the same quarter you deploy. This is the driver of both rapid commercial growth and funding. The Coalition for Health AI launched PULSE, a national initiative to help public health agencies evaluate and scale GenAI, on July 16. OpenAI and Anthropic donated 10 enterprise licenses covering up to 2,000 practitioners in total, roughly 200 in each of 10 state, local, tribal, or territorial jurisdictions, with Accenture managing onboarding and turning the pilots into shared playbooks. Applications close August 7; pilots begin this fall, and the playbooks are expected to go public in 2027. My Take
Public health often can't buy frontier AI at market rates, so the AI labs are donating. Genuinely useful, but also a smart distribution strategy. The key will be post‑pilot when the invoices begin. Ideally the ROI is evident by then and the tools fund themselves. Bunkerhill Health closed a $25 million Series B led by Khosla Ventures on July 16, bringing total funding to $55 million, with Sequoia, Felicis, Optum Ventures, and Y Combinator on the cap table. Its Carebricks platform lets health systems build and govern their own production AI agents rather than buy point solutions. The platform runs today at Cleveland Clinic, the University of Texas Medical Branch, and Intermountain Health. At UTMB, 22 agents are live; the company says its nephrology triage agent there cut specialist wait times by more than half. My Take
The round is modest by this year's standards. Still, the client list is impressive. Big systems like Cleveland Clinic and Intermountain building their own agents tell us where that end of the market is going. Open question is where smaller systems land. |
|
2
Research Studies
Managing diabetic kidney disease means juggling patient history, shifting biomarkers, and evolving treatment guidelines. Researchers grounded a locally deployed model in a curated corpus of clinical guidelines, so it retrieves the relevant rule before answering, then paired it with a cloud‑based reasoning engine. This approach was tested retrospectively on 267 multi‑center patient cases with blinded review by 12 independent physicians. It beat unaugmented models on guideline‑concordant recommendations and substantially reduced safety‑critical errors, especially medication contraindications tied to a patient's kidney function. My Take
The safety gain came from the retrieval layer (vs. a bigger model, more training data, etc.). This study highlights the value in grounding any model in your own guidelines, or the guidelines you trust, before you scale. An imaging‑free machine learning model predicts 30‑day mortality in intracerebral hemorrhage patients using data already gathered in the first 24 hours of ICU care: labs, vital signs, age, and Glasgow Coma Scale scores, but no brain imaging. Trained on 1,034 patients and tested on 444 more, it reached an AUC of 0.859, where 1.0 is perfect and 0.5 is a coin flip, and held at 0.811 on 339 patients treated at the same hospital a decade earlier. My Take
An AI‑predicted score built from data already collected at the bedside is another example of an easy‑to‑see value‑add from AI. |
|
1
Key Insight
AI companies are now a commercial determinant of health
Epidemiologists have a name for how industries shape health outcomes: “commercial determinants of health”. Tobacco earned the label first; food and pharma followed. A BMJ Analysis this week argues AI companies now belong on that list. The authors' case: firms like OpenAI and Anthropic are no longer just tool vendors but also a fast‑growing commercial force with real influence over health policy, regulation, and the health information people actually see, and GenAI makes the classic influence playbook more targeted, more adaptive, and harder to detect. Regulators and health systems aren't keeping pace. WHO's European office warned this month that roughly two‑thirds of its 53 member states already run AI diagnostics while only 8% have a health‑specific AI strategy, and fewer than half have even checked whether their existing laws cover it. A global survey of 1,823 clinicians, run by the AI scribe company Heidi, found that 83% of those using AI at work do so without employer guidance. Takeaway
AI companies are getting organized. Health systems around the world have work to do. The good thing is health leaders already know how to manage commercial influence: disclosure, evidence requirements, procurement discipline; pharma got that treatment decades ago. Today, hard to argue against getting even smarter about the AI companies. |
|
Know someone who'd find this useful? Share |
New buzzword? AI companies as "Commercial Determinants of Health"
3 signals, 2 studies, 1 insight for health leaders
Jul 24, 2026
•
3 min read