AIHealthTech Insider: Issue #124
October 26, 2026
Illustrations in this issue are AI-generated.
Hello,
A diagnosis can take several steps: asking questions, ordering tests and reconsidering an early answer.
Microsoft’s latest research tests AI on that process.
This week: what it got right, Google’s approach to personal health support and a protein-design challenge that goes beyond binding.
For education, not medical advice.
☕ Quick question before we dive in 👇

What would you ask your smartwatch if it could talk?
👇 Later in this issue: Can an AI-designed protein activate the receptor targeted by Ozempic?
🔬 The Big Story
Microsoft’s AI worked through difficult diagnoses
Microsoft’s MAI-DxO brings together AI agents that ask questions, choose tests and challenge each other’s conclusions.

In research published in Science, it solved 83% of 304 complex medical cases drawn from the New England Journal of Medicine. Physicians achieved 36% with reference materials and 22% without them.
The system received information gradually, rather than seeing the complete case immediately.
Why it matters: This tests how AI investigates a problem, beyond simply choosing an answer.
The honest part: These were case-based evaluations. They don’t establish better outcomes for patients. Prospective testing with Beth Israel Deaconess Medical Center is planned.
New here? Join 1,500+ readers getting one carefully sourced AI-healthcare briefing every Monday—written in plain English, without the hype.
🤎 Sponsored
Let world-class chefs answer your cravings.
Tired of the same meal prep on repeat? CookUnity can help. Because they connect you with 500+ meals handcrafted by over 130 world-class chefs, including Michelin-starred chefs and Food Network stars.
Explore 60+ different cuisines and diets, and choose dishes that fit your cravings and preferences. Your meals arrive fresh and ready to heat in minutes, so no prep or cleanup.
Whether you're in the mood for a comforting classic or a new flavor, there's always something new to discover because the menu rotates weekly, and their chefs are always creating new recipes.
Explore the menu available in your area and build a weekly plan you can skip, pause, or adjust anytime.
⚡ Quick Hits
🔸Google’s health agent divides the work three ways
Google’s Personal Health Agent combines a data analyst, health expert and coach to interpret wearable data and health records.

Published in Nature, the research evaluated answers, analysis and coaching conversations.
The honest part: It hasn’t demonstrated lasting health improvements. Reviewers also identified potentially harmful content in some generated summaries.
🔸AI-assisted TB screening reaches more clinics
Google reports that partner Nexus AI has supported roughly 25,000 chest X-rays across 40 health facilities.

The technology flags possible tuberculosis for further investigation.
The honest part: A screening flag isn’t a diagnosis. Confirmatory testing remains necessary.
Know someone in healthcare, medicine, or AI? Forward this to them. They can subscribe free here →
🧠 Brain in Brief
One extra school year, fewer dementia diagnoses later
An NIH-supported study examined a UK policy that raised the school-leaving age from 15 to 16.

Among more than 100,000 people, the additional schooling was linked to fewer dementia diagnoses, particularly among those with higher genetic risk.
The honest part: Follow-up extended only to around age 65. This doesn’t establish that taking a class later in life prevents dementia.
🤎 Sponsored
Your Fall Glow-Up Comes in Two Cups
Pique’s Radiant Skin Duo pairs antioxidant-rich Sun Goddess Matcha with B·T Fountain, a beauty electrolyte for hydrated, plump-looking skin. In a trial of B·T Fountain’s ceramides, hydration improved 37% and elasticity 36% by day 60. Subscribe for 20% off for life, backed by a 90-day money-back guarantee.
🌿 Worth Knowing
Can an AI-designed protein switch on a drug target?
Anthropic and Adaptyv’s third protein-design challenge asks for proteins that bind to—and activate—GLP-1R, the receptor targeted by semaglutide, sold as Ozempic and Wegovy.

Selected designs will undergo laboratory testing, including checks for receptor activation. The results will be shared publicly.
The honest part: These are experimental candidates. Laboratory activity wouldn’t establish that they are safe or effective medicines.
One useful question this week
When a health headline catches your attention, ask:
“Did this improve someone’s health—or is that what researchers need to test next?”
We’ll keep making that distinction clear.
Know someone who enjoys health research without the jargon? Forward this issue.





