In my senior year of college, I helped teach a freshman poetry workshop. Twenty-year-old Adam unironically coached eighteen-year-olds to hone their observations on sex and death with an unearned confidence. Metaphors of passionate fires and endless darkness prevailed. Last week's clinic covered the same subjects, but thirty years and a medical degree shape metaphors.
Hospice came up several times, mostly with patients who are not transplant candidates and do not want dialysis. None of it was urgent, and in place of visions of perpetual night or Valhalla, we were working through tactics. What do you say to a child who is pushing for dialysis? When do you mention to your spouse that you don't want aggressive care? Helping patients find the words, despite loud family members who feel they get a vote, was the conversation.
And then there were two older men, each planning trips with their lady friends, both requesting Viagra. A kidney doctor is not the front-line prescriber for this. But they asked me, and what followed was not a conversation about desire or the body's betrayal. It was a found poem about blood pressure, heart medication interactions, and the four-hour warning, acknowledged with the nod you give a flight attendant after the emergency exit briefing. Somewhere on an interstate this summer there will be two late-model cars, yacht rock, lady friends, and prescriptions that mitigate age.
The first-year students had all the wonder and none of the facts. My patients have all the facts and none of the wonder. I am not sure which version is the more honest poem. Thirty years ago I helped teenagers reach for language big enough to hold what they had not yet felt. Last week I helped older adults find the small words to describe decades of feelings. The mystery is not gone. We just use different metaphors.
I have moved all the older articles to a new hosting provider and updated the search and index at http://www.whatadamisreading.com. All supporting AI summaries, images, audio, etc. will be hosted on the new provider since some organizations block personal Google Drive accounts and Claude.AI artifacts. Let me know if you have any problems.
The Google Gemini Notebook AI-generated podcast version of this week's newsletter.
https://wair.ajwein.com/wair-podcast-2026-07-27-v1
Science and Technology Trends
There are many ways to tinker with proteins. Making proteins more active or more useful is the hard part, and typically requires years of trial and error. Researchers at UC Berkeley used an AI protein-design model, guided by the evolutionary patterns of thousands of natural proteins, to redesign an enzyme in days rather than months. The enzyme was TnpB, the primitive ancestor of CRISPR, so what they built is a working gene-editing system that does not exist in nature, with a couple of versions that edit more effectively. It is a small example of how AI is compressing the time from question to insight in biology.
https://wair.ajwein.com/syntnpb-ai-nucleases-2026-07-23-v1
Last week we had kidney (cells) in space. This week, we have kidneys on ice. Researchers cooled pig kidneys to about −4 °C without letting ice form, stored them for up to 72 hours, and then transplanted them back into the same pigs. The animals lived on that one thawed, working kidney for months. In fact, the kidneys recovered mostly normal function in about 10 days. This is not a peer-reviewed or published dataset (it is more of a tech-focused piece about the company commercializing the device). However, finding more effective ways to store and transport human organs is a big deal. Several technologies are being developed for prolonged organ storage and transport; supercooling extends organ viability from hours to days, a step toward the kind of shelf life we see in reproductive cell banking.
MIT Technology Review, July 23, 2026
https://wair.ajwein.com/supercooled-kidneys-2026-07-24-v2
Last week the White House Office of Science and Technology Policy released Science: A New Golden Age. The report models itself on Vannevar Bush's 1945 report to President Truman, "Science, The Endless Frontier." The new manifesto proposes to redirect hundreds of billions in federal research funding away from universities and toward industry, artificial intelligence, and individual scientists. This is the intellectual scaffolding for how the administration wants to restructure a roughly $200 billion federal R&D enterprise, with direct consequences for NIH, NSF, and biomedical research. The caveat, raised by supporters and critics alike in STAT's coverage, is the distance between what the report preaches and what the same administration is actually doing.
https://wair.ajwein.com/golden-age-science-2026-07-24-v1
Anti-Anti-Science
Katelyn Jetelina, PhD, MPH, leads Your Local Epidemiologist, an independent public health newsletter that employs physicians, epidemiologists, immunologists, and nutritionists. This week, she covers three concurrent infectious disease stories: a COVID uptick in the western U.S., a measles case count surpassing 2025's full year total within the first seven months of 2026, and a Cyclospora outbreak that has sickened over 11,000 Americans from contaminated lettuce. These outbreaks are symptoms of a system simultaneously de-staffed, de-funded, and de-legitimized. The circularity is the problem - weaker infrastructure produces worse outbreaks that further tax the system that remains.
https://yourlocalepidemiologist.substack.com/p/covid-19-waking-up-cyclospora-marches
https://wair.ajwein.com/yle-dose-infectious-disease-roundup-2026-07-21-v1
On July 23 and 24, the FDA's Pharmacy Compounding Advisory Committee voted to recommend six peptides for the Section 503A bulk drug substances list: BPC-157, KPV, TB-500, MOTS-c, Epitalon, and Semax. It rejected a seventh, emideltide, for insufficient evidence. This means compounding pharmacies, not just gray-market suppliers, can source these components and presumably deliver a purer and more consistent product.
Multi-source review of the panel's decision:
https://wair.ajwein.com/fda-peptides-compounding-2026-07-26-v1
It's important to remember that the data on the actual clinical utility, safety, and efficacy of these peptides is still very much in doubt. I've reviewed peptides on numerous occasions, going back to the summer of 2025.
AI Impact
A large group of academic medical researchers built NOHARM, a benchmark that measures how often AI-generated medical advice could actually harm a patient, and found the potential for severe harm in up to 25% of cases for the worst-performing AI models (the purpose-built clinical tools landed closer to 3 to 5%). Interestingly, more than 80% of the severe errors were errors of omission: the LLMs failing to offer return precautions, safety-netting, "come back if X," follow-up timing, and patient education. These kinds of instructions are a hedge against diagnostic uncertainty and patient inconsistency. The "come back if X" is a backstop against wrong diagnoses or ineffective treatments.
In a companion study of 101 randomized attending physicians, AI assistance improved doctors' clinical management plans. Yet the doctors often ignored good AI suggestions. Keep in mind that these are potential harms as scored by specialist reviewers on review checklists, not observed clinical outcomes, and the paper is not yet peer-reviewed.
The emerging pattern is more unsettling than "physicians plus AI beat physicians alone," even though this study demonstrated this outcome. In this study, the human-plus-AI team still lost to the AI working alone, which was amplified by physicians discarding AI recommendations. The same group's earlier work found similar results. The AI+doctor beats the unaided physician, but it underperforms when the AI suggestions are ignored, and often is not as good as the AI by itself. Since we have not yet licensed AI to practice medicine, we'll need to figure out how to get doctors and AI working together effectively.
https://wair.ajwein.com/noharm-benchmark-2026-07-23-v2
I learned about NIMBY (not in my backyard) in the 1980s, when energy companies were having difficulty finding appropriate locations for nuclear power plants. More recently, the acronym is being applied to data centers. Gallup recently published the first opinion survey on AI infrastructure, drawing from a large sample of American citizens. The topline: 71% of American adults oppose building an AI data center in their area.
A deeper dive suggests something more complicated than simple opposition. People who oppose data centers cite water, energy, and land consumption, as well as quality-of-life concerns and increases in utility bills. But the survey doesn't explore whether respondents understand what data centers actually do, whether the phrase "AI data center" triggers anxieties that "data center" alone would not, or whether people have considered AI's strategic importance as a competitive advantage in the global marketplace. Supporters, for their part, overwhelmingly cited job creation as the upside.
The framing is incomplete. The question isn't whether people want a data center next door (who wants anything next door). It would be interesting to see how people respond to a balanced discussion of the trade-offs involved in expanding AI infrastructure.
https://wair.ajwein.com/gallup-data-centers-2026-07-22-v1
The J&E Random Kidney Facts of the Week (JERKFoW!)
Two loyal readers suggested I salt the newsletter with random kidney facts for the nephrologic edification of the readers. Be edified, friends!
One of the most common questions I get when patients see their lab results is about their sodium. It often surprises people to learn that the number labeled "sodium" isn't really a measure of salt. It's a measure of how watered-down your blood is. When it runs low, a condition called hyponatremia, the problem is rarely too little salt; it's too much water, more than the kidneys can excrete. Hyponatremia is the single most common electrolyte abnormality in hospitalized patients, and a fair amount of my specialty's after-hours phone time goes to coaxing salt and water back into balance, sometimes with medications that push the kidneys to dump the excess water into the urine.
And yes, you can die from drinking too much water, but most people do not have diseases or behaviors that lead to lethal levels. Healthy kidneys can excrete close to a liter of water an hour, so ordinary drinking, even the hydrate-all-day, big-bottle kind, never comes close to overwhelming them, and frequent bathroom trips are a good disincentive to continued drinking. In an infamous example, a 28-year-old California woman drank more than 1.5 gallons during a 2007 "Hold Your Wee for a Wii" radio contest to win a Nintendo console. She died of water intoxication from quickly ingesting an enormous volume of water. Bottom line: Unless otherwise instructed by your physician, drink when you're thirsty; your kidneys will handle the rest.
Here is a recap of the Wee for a Wii case:
Things I learned this week
Suppose Evolution had played out differently. We could have had eight fingers on each hand as opposed to five. Last week, I found an interesting popular science article about why, at least, we think most creatures on the planet have five-fingered hands, paws, and paddles. While the why is not 100% clear, it is clear that the evolution of the genes encoding 5 digits solidified in the genomes of most earth-bound creatures about 360 million years ago. Imagine a world of keyboards and other equipment accommodating eight fingers on each hand. While I can't explain the Imperial Counting System, I think the metric system would have had significant competition for base-16 numerical systems had an 8-digit evolutionary reality played out.
https://www.popsci.com/science/why-five-fingers-toes/
https://wair.ajwein.com/five-fingers-2026-07-26-v1
This week I learned that the NY State municipality of Salamanca recently signed a contract with Realbotix to integrate a seated, lifelike humanoid robots into a sponsored STEM curriculum that is intended to serve underserved students in the district's technology programs. Fear, uncertainty, and doubt ensued. Community members and the New York State United Teachers Union are criticizing and questioning the safe and effective use of AI in the classroom. Moreover, it turns out that the company that made this robot is also associated with the deployment of lifelike sex robots, which this robot was not, but became another source of consternation for all involved. The Ground News headline roundup captures the range of reasonable to absurd, including "School sparks debate after buying humanoid robot teacher," and my favorite, "Sex doll company strikes bizarre deal to put its robot in high school classrooms." Clearly some headlines are designed to inform, others are designed to capture clicks. I wonder what sort of teenage poetry this robot teacher could have inspired.
Photo of the Realbotix not-sex-robot teacher:
AI art of the week
A visual mashup of topics from the newsletter, and an exercise to see how various LLMs interpret the prompt. I use an LLM to summarize the newsletter, suggest prompts, and generate images with different LLMs.
Gallery:
A panel of four Norman Rockwell-style Saturday Evening Post covers, arranged in a 2×2 grid. Each cover faithfully reproduces the classic Post layout: the "THE SATURDAY EVENING POST" masthead in its distinctive serif font across the top, a "5¢" price badge in the upper right corner, the date "July 27, 2026" beneath the masthead, and article headlines in small red and black Post house style typography tucked into the margins around the edges of the central illustration, occasionally overlapping the artwork.
Full Prompt:
https://wair.ajwein.com/wair-art-prompt-rockwell-2026-07-27-v1
Clean hands and sharp minds,
Adam
Relive all the past thrills and excitement - The What Adam is Reading Archive
http://www.whatadamisreading.com/