Bonus content for paid subscribers: This post comes with my complete critical appraisal of the JAMA Perspective (PDF), documenting every study, citation, and conflict of interest.
Who would have thought I’d see the day when AI slop showed up in one of the most respected medical journals in the world?
I was wrapping up patient hours when an email notification caught my eye: “Will Autonomous AI Exceed AI-Aided Physicians as the Best Medical Care?” (Emanuel, Baker-Butler, Khosla & Khosla, JAMA 2026) Among its authors are Dr. Ezekiel Emanuel, venture capitalist Vinod Khosla, and his son Neal Khosla, CEO of an AI-enabled health company.
Their argument: AI working alone already matches or beats physicians at five cognitive tasks, from taking a history to managing chronic disease. And once AI is better at a task, keeping a physician in the loop makes results worse. Their forecast is autonomous AI in real-world workflows by 2030.
By “AI slop,” I don’t mean the prose. I mean AI hype dressed up as evidence.
I find this piece deeply troubling, and not because I’m afraid AI will replace me. Not one cited study measures what happens to patients under autonomous AI care. No mortality. No major adverse cardiac events. No hospitalizations. No adverse events. Add significant industry ties among the authors, and you have a Perspective that will gaslight a great many physicians.
How I wrote this, and why it matters
I’m not an AI skeptic. I build AI tools for cardiovascular risk assessment, and I teach physicians to build AI-native practices. So here’s exactly how this piece came together, because the process is the argument.
It started with my own recorded critique, which is the original work here. You'll find the transcript at the end of this post, edited for clarity and corrected where I misspoke. Then Claude in Chrome produced a structured critical appraisal of the article’s claims, studies, citations, and conflicts of interest, which I checked against my own read. Finally, a Claude project I use for my newsletters turned the appraisal and the transcript of my recording into this article, and generated a summary, which I edited.
That’s a physician-led, AI-assisted workflow, the very model this Perspective says holds medicine back. AI did the retrieval and cross-checking. I made the judgments calls and I’m accountable for every claim and the narrative arc of my opinion. The difference between AI-assisted work and AI slop isn’t whether AI touched it. It’s whether someone accountable stands behind it, and whether the evidence holds up.
What “AI beats doctors” rests on
The authors say they reviewed “all published articles on AI in medicine” since January 2024 and found a “preponderance” of studies on their side. No search strategy, no databases, no inclusion criteria, no risk-of-bias assessment. That’s not a preponderance. It’s a laundry list.
Google’s AMIE study was a text-chat simulation with actors playing patients, on an interface primary care physicians don’t normally use. Its striking 97% versus 50% figures are how often raters judged performance favorably, not measured accuracy.
Microsoft’s MAI-DxO study, still a preprint, used NEJM teaching puzzles and barred physicians from textbooks, the internet, and colleagues. None of us practice that way.
And the “AI” in the article’s showcase chronic-disease trial? It wasn’t generative AI at all.
It isn’t the machine that is coming after the doctors. Subscribe to discover:
The 32-patient trial doing the heaviest lifting, and why it’s not an AI application but physician-led automation
The specialty excluded outright, and the author who made a failed public prediction about it
The 106,000-woman randomized trial cited as an example of bias, which actually shows the hybrid winning
The investments and family ties behind the byline
A citation that says the opposite of the claim it supports
What I’m asking JAMA to do
My original critique in my own words, plus the complete critical appraisal
The full breakdown, my original critique in my own words, and the complete critical appraisal PDF are for paid subscribers. If you want the evidence and not the hype, upgrade to read the rest.