Two newsletters, one breakthrough a week

Both live on LinkedIn, 2,000+ readers between them. Each edition opens with the 30-second version, then the detail. Scroll a column for the latest editions; the links open the article on LinkedIn.

AI in Healthcare

This week's breakthrough, read by someone who has worn the white coat.

Every week I pick the one development in healthcare AI that actually changes something — a study, a deployment, a regulator's move — and explain it in the 30-second version first, then the detail. Researched with AI assistance, reviewed and published by me.

Weekly, Fridays, on LinkedIn.
Monthly edition: the month's buyers, regulators and evidence in one read.

Subscribe on LinkedIn →

Oct 23, 2026

A Virtual Panel of Doctors Heads to the ER

Microsoft's MAI-DxO solved 83% of complex diagnostic cases in a Science study; generalist physicians working alone with references solved 36%.
It works like a panel of doctors: several AI agents question the patient, order tests and debate the diagnosis. The gains shrink on emergency cases.
What matters next: a real-world evaluation in Boston emergency and urgent care, alongside human doctors.

Read on LinkedIn →

Oct 16, 2026

271 Million Questions Doctors Asked an AI

JAMA published the largest look yet at how clinicians use AI: 952,008 US clinicians, 271 million questions, two and a half years.
They mostly ask about treatment. Diagnosis — the task nearly every benchmark tests — was 13.9% of queries.
Physicians within a decade of training are 24% of the workforce and sent 62% of the questions. JAMA AI launches in early 2027.

Read on LinkedIn →

Oct 9, 2026

Can Your Audit Log Be Trusted?

Epic found flaws in certain MyChart configurations that could let someone view patient records without the access appearing in an audit trail — by pointing an AI at its own software.
Epic pulled engineers off product development for roughly six weeks to patch. Close to unheard of in enterprise health IT. MyChart holds more than 320 million patient records.
The uncomfortable part: finding flaws is now easy. Patching them is the bottleneck.

Read on LinkedIn →

Oct 3, 2026

September 2026 — The Month the Buyers Arrived (monthly edition)

September was the month the purchasers showed themselves: a major insurer bought frontier AI, a federal agency committed $62.7 million, the EU named healthcare a priority sector, the White House ordered an AI front door to federal health services.
The FDA now lists over 1,600 authorized AI-enabled medical devices — and in the same month refused to loosen the rules for radiology AI.
The evidence question got sharper, not softer: an advisory panel split 6–4 on the first multi-cancer blood test; case managers beat AI at predicting discharge when it mattered most. Money and institutions moved while the proof stayed contested.

Read on LinkedIn →

AI in Practice

Real-world AI adoption: news, success stories and use cases across every industry and daily life.

The optimistic newsletter — but specific. One breakthrough a week that you can actually use or point to, with the numbers and the caveats, and none of the hype.

Weekly, on LinkedIn.
Monthly deep-dive: the month's models, prices and guardrails in one read.

Subscribe on LinkedIn →

Oct 25, 2026

Why Is OpenAI Suddenly on Its Rivals' Clouds?

On October 21, OpenAI announced its models and Codex tools are coming to Google Cloud — which puts them on all three major US clouds.
Microsoft's crown jewel now sells through Google; Google's biggest AI rival now runs on Google's infrastructure. Everyone signed anyway.
The strategic read: the frontier model just unbundled from the cloud. Model choice and cloud choice are now two separate decisions — and separate negotiations. Your leverage went up this week, if you use it.

Read on LinkedIn →

Oct 19, 2026

The Week's Best AI Fits in Your Pocket and Can't Read Your Chats

Google DeepMind and GovTech Singapore unveiled an on-device scam detector: a fine-tuned Gemma model that spots manipulative conversations on your phone and warns you mid-chat.
Everything runs locally, zero cloud egress — neither Google nor the Singapore government can read a message through it. Scams are more than half of Singapore's reported crimes, over S$4 billion lost since 2019.
Same week, the loud news ran the other way: a $1 trillion AI chip selloff and new probes. The useful story was the quiet one.

Read on LinkedIn →

Oct 11, 2026

722 Proofs in One Drop: AI Now Outruns the People Who Check It

On October 6, OpenAI published 722 mathematical manuscripts from an unreleased internal model to a public GitHub repo: 372 result families across 17 fields, including a claimed proof of the Unique Games Conjecture.
Nearly every result traces to a single prompt handed to one agent, at roughly three hours of compute per problem. Many ship with machine-checkable Lean proofs; none has passed peer review, and the humans who could review them are refusing or overwhelmed.
The quiet part, via Scott Aaronson: frontier labs have begun testing whether math-capable models can break real cryptographic primitives.

Read on LinkedIn →

Oct 6, 2026

September 2026 — The Month Saying No Became a Business Model (monthly deep-dive)

September opened with OpenAI shipping its first Critical-rated model behind gates, and closed with it shelving the successor entirely for deception and scope failures.
Underneath the restraint arc, the money moved: Anthropic's IPO filing revealed $518 billion in infrastructure commitments; OpenAI pushed its own listing to 2027.
September 22 collapsed frontier pricing to a ten-cent floor nobody predicted in January. The quiet story: the rival labs spent the month building shared guardrails, including a near-binding pact to stress-test each other's models.

Read on LinkedIn →