🗞️ Opening Line
Happy Friday! AI just beat experienced physicians at diagnosing patients in a real emergency room. JPMorgan put a $2.5 billion number on its AI returns. And Oracle is cutting 30,000 jobs to fund its next infrastructure bet. Let's get into it.
💡 The Big Update
AI Just Outperformed Doctors at Diagnosing Patients. This One Is Hard to Dismiss.
A study published this week in Science — one of the most prestigious peer-reviewed journals in the world — found that an OpenAI reasoning model outperformed experienced physicians at diagnosing and managing patient care in a real emergency department setting.
The research came from Harvard Medical School and Beth Israel Deaconess Medical Center. The AI worked only from electronic health records — the same data doctors use — and outperformed physicians on both diagnostic accuracy and care management decisions across a Boston ER dataset.
This isn't a benchmark. It isn't a controlled lab experiment with perfect inputs. It's a real hospital, real patient records, and real physicians as the comparison group. And the AI won.
To be clear: this doesn't mean AI is replacing doctors tomorrow. The study reflects one model on one dataset in one setting. Clinical deployment involves regulatory approval, liability, workflow integration, and patient trust — none of which a research paper resolves. The researchers themselves were careful to frame it as a milestone, not a deployment recommendation.
But it is a milestone. The question "can AI do what a doctor does?" just got a much more serious answer than it had a week ago. Healthcare is one of the most complex, high-stakes domains humans have ever tried to systematize. If AI is competitive here, the list of domains where it isn't is getting shorter fast.
⚡ Quick Commits
JPMorgan Put a Number on AI Returns — $2.5 Billion This Year JPMorgan Chase formally reclassified AI from experimental R&D to core infrastructure this week, with a 2026 technology budget of $19.8 billion and 2,000 staff dedicated to AI development. The bank expects AI to generate $2.5 billion in annual value through efficiency gains and revenue growth. Models are already scanning over $10 trillion in daily transactions. For anyone still waiting on proof that enterprise AI pays off — this is the clearest number yet from one of the world's largest financial institutions. Crescendo AI
Oracle Is Cutting 30,000 Jobs to Fund AI Infrastructure Oracle announced plans to cut between 20,000 and 30,000 employees to redirect $8–10 billion toward AI infrastructure. It's the same pattern we've seen from Snap, Meta, and others — workforce reduction as a funding mechanism for compute investment. The AI infrastructure buildout isn't slowing down. It's being financed, in part, by the people it's replacing. Crescendo AI
🛠️ The Tool Drop
Whoop Sticking with the healthcare AI theme — Whoop just announced it's adding AI-powered health guidance and on-demand video consultations with licensed clinicians directly inside the app for US users. It's also adding electronic health records integration. For anyone already wearing a fitness tracker, Whoop is quietly becoming something closer to a continuous health monitoring platform. The line between consumer wellness tech and actual healthcare infrastructure is blurring fast. whoop.com
🎯 The So What
The ER study, JPMorgan's $2.5 billion ROI, Oracle's 30,000 cuts. Three stories that tell the same story from three different angles.
AI is moving from the edges of industries into the center of them. In healthcare. In finance. In enterprise infrastructure. The institutions that have the most to lose from disruption are the ones moving fastest — because they can read the numbers.
For the rest of us, the takeaway is still the same one we come back to every week: the gap between organizations that are building AI into their operations and those that aren't is widening. The ER study just made that gap feel a lot more real.
Have a great weekend.
— The Changelog
Not subscribed yet? Get The Changelog free, three times a week. Subscribe here
Know someone who'd find this useful? Forward it along.