First, a re-cap of our Alexandr Wang interview — his first real sit-down in eleven months — and what it actually revealed about Meta’s AI play. Wang seemed nervous hashing out the strategy in the studio, and we both keep circling the same puzzle: Meta has endless compute and top talent in Nat Friedman and Daniel Gross, so why does the model still feel underwhelming?

Subscribe now

We get into Eric Schmidt getting booed off a commencement stage at the University of Arizona, which becomes a longer conversation about the generational fury aimed at AI. Everyone Kylie’s age seems to hate it, but is it due to misinformation or legitimate anger about jobs and data centers? Ashlee admits he’s more confused by this moment than anything he’s covered in tech: the predicted Wall Street collapse hasn’t come, the models keep getting better, and the valuations still make no sense.

Then, the news that broke minutes before we hit record: OpenAI won the Musk lawsuit on statute-of-limitations grounds. We dig into whether OpenAI’s shift from open-source nonprofit to for-profit was an original sin or just the only way to pay for the compute. Also, Ashlee’s texts with Sam Altman being part of discovery?!

In more Musk news, Bloomberg reported that Musk’s xAI stiffed staff on the $420 they were promised for feeding their tax returns into Grok. One host would decidedly not trust a chatbot with their financials, and the other already has. We also get into the strange new bedfellows: SpaceX selling compute to Anthropic, a company Musk has long been philosophically against.

We even take you behind the scenes at Core Memory — so study up and watch our latest videos on Phantom Neuro and Starfront Observatories. Consider this your homework on mind-controlled arms and galaxy photography. Essay due on our desk by morning.

Don’t forget you could win tickets to the Weezer tour by leaving an amazing review for our podcast wherever you listen.

Subscribe now

The Core Memory podcast is on all major platforms and on our YouTube channel over here. If you like the show, please leave a review and tell your friends.

OUR SPONSORS

SendCutSend

Do you make stuff? Do you need metal parts fast and believe in truth and justice? Then head on over to SendCutSend where you’ll get a 15 percent discount thanks to Core Memory on whatever you’re trying to build. We believe in you.

Brex

The Core Memory podcast is also sponsored by Brex, the intelligent finance platform built to help companies spend smarter and move faster.

Did we go to Texas, find a telescope ranch and then obtain an entire nebula in Brex’s honor? Oh yes, we did.

We run on Brex and so should you. Learn more about Brex right here.

It’s with great pleasure that we present this excerpt from The Impossible Factory by Josh Dean. It’s a tremendous, new book about Kelly Johnson and Lockheed Skunk Works.

Copyright 2026 by Josh Dean. Published by Dutton, an imprint of Penguin Random House.

Kelly Johnson knew that engineers, especially those in defense contracting, had a responsibility to understand and predict the market. No commercial enterprise that lives at the bleeding edge of technology can survive for long if it doesn’t anticipate the needs of its buyers far in advance. So, Kelly was constantly talking to his contacts in the Defense Department and in the national intelligence establishment about the global chessboard and the challenges that lay ahead.

He caught wind of a “desperate need” for a new type of American aircraft before anyone asked him for it—one that “could safely fly over the USSR” and bring back critical information on Russia’s missile capability and other details about its defenses and military infrastructure.

Subscribe now

On May Day 1954, the Soviets unveiled their latest nuclear bomber—the Myasishchev M‑4 “Hammer”—which soared low over Red Square, creating quite a stir in Washington, especially because it hadn’t even been a year since the USSR had detonated the world’s first hydrogen bomb.

Top officials—including, if not especially, President Dwight Eisenhower—were particularly worried about the Soviet Union’s strategic bombers, like the Hammer, which could carry nuclear weapons and which U.S. military and intelligence leaders knew almost nothing about: a type of plane that would potentially allow for a Pearl Harbor–style sneak attack, but far worse.

In the spring of 1954, Eisenhower asked James Killian, president of MIT, to form a committee to make recommendations for how the United States could leverage its tremendous base of scientific and technological firepower to determine what the Soviet military was capable of—and as a result, how much danger America was truly in.

A subcommittee was told to “find ways to increase the number of hard facts upon which our intelligence estimates are based, to provide better strategic warning, to minimize surprise in the kind of attack, and to reduce the danger of gross overestimation or gross underestimation of the threat.” No pressure there.

The United States and the Soviet Union were still in the early days of a nuclear arms race, and any edge in that race created leverage in the battle for global supremacy. But there was a fundamental imbalance when it came to intelligence gathering. As a free and open society, the United States was susceptible to on‑the‑ground spying. But the Soviet Union, being a closed, authoritarian state, was nearly impossible to infiltrate with spies.

To make up for that, the United States had to be creative. It would need to use science and technology to out‑spy the Russians. And if American spies couldn’t get into the Soviet Union to gather intel, they’d have to fly over it, which presented its own challenges. Like being shot down.

The design challenge facing Kelly Johnson, then, was daunting: To safely overfly the Soviet Union and take high‑quality photos undetected required a plane that could fly more than four thousand miles without refueling, and reach at least 70,000 feet—beyond the reach of Soviet air defenses and so high that it wouldn’t create vapor trails, thus revealing itself.

In short, this plane would need to be extremely light, while carrying an array of the most advanced cameras, sensors, and navigational gear available.

The need was also urgent. Existentially so.

That March, Kelly submitted Lockheed Report #9732 for an ultralight, high‑flying surveillance plane with an enormous wingspan to the Air Force.

The pitch was radical in that the plane Kelly was proposing had no landing gear, to save weight. Being as light as possible is mission critical for flying high, so Kelly was looking for weight savings wherever he could find it and decided that his plane would drop its gear upon takeoff and land on a reinforced belly.

The pitch was not a hit. Kelly received a letter from the Air Force declining the proposal “on the basis that [the concept] was too unusual.”

But one Air Force official loved the idea: Trevor Gardner, the “technologically evangelical” assistant secretary for research and development. Gardner knew of a different buyer who might be interested and summoned Kelly to Washington in November for an urgent meeting.

On November 19, Kelly met with a group of officers, engineers, and scientists, and endured a grilling that reminded him of his college days.

Shortly thereafter, top officials took the proposal to Eisenhower in person, because the president feared leaks and the subject was considered too highly classified to be put in a written report.

Eisenhower approved the plan, with a stipulation. “It should be handled in an unconventional way so that it would not become entangled in the bureaucracy of the Defense Department.”

It was given instead to the CIA.

In advance of his trip to Washington, Kelly had been warned by his bosses not to commit to anything and he worried that he might have to take a leave of absence from his job at Lockheed proper to take on this project.

Lockheed’s capacity was straining, especially in engineering. Still, when Kelly met Robert Gross and Hall Hibbard—the only two men he was cleared to tell—he told them that this was a job Lockheed had to take. And that secrecy demanded that he run the entire program, from design to manufacture, in his Skunk Works.

The two bosses heard him out and agreed.

This was, arguably, the biggest single moment in the history of the Lockheed Skunk Works, in that Kelly now had approval for something more than an experimental design shop. He was given the green light to run his own production, too. Which meant that he wasn’t just building prototypes. He would oversee full production of all planes built for the program.

What’s more, the government was willing to hand him unprecedented control. Lockheed was taking “full responsibility for the design, mockup, building, secret testing, and field maintenance of this unorthodox vehicle.”

Kelly spent two days refining the concept himself, then summoned five key Skunks to his office for a meeting.

He looked at the assembled talent and spoke of a new program, one more secret than anything any of them had ever worked on, and then, without revealing any details about what they’d actually be doing, asked if they were willing to commit eighteen months to such a project.

All five said that, yes, they absolutely would. And then Kelly leveled with them. He’d sold the CIA a high‑altitude reconnaissance airplane. They could have a few days to wrap up their current work but should be ready to go full bore on Monday, December 2. It was time to make history.

—

Kelly’s reconstituted Skunk Works began with twenty‑five engineers, with his trusted shop man, Art Viereck, in charge of production. Kelly assigned Ed Baldwin to handle the traditional three‑view drawing.

Four days later, Baldwin had the first drawings completed, and by December 10 the basic design was frozen. It was, more or less, the configuration that would go into production. Which is fairly astounding to consider once you know what the plane Kelly laid out and Baldy sketched would become.

Shortly after beginning, Kelly prepared a twenty‑three‑page report for the CIA with his updated thoughts on the plane his Skunk Works would build. Among them, that the Angel, as he was calling it, would have a maximum speed of Mach 0.8 (460 knots) in level flight, with a ceiling of 73,100 feet—an absurd altitude that had only been reached at this point by research balloons and a few highly experimental one‑off aircraft.

He promised to have the first plane flying by August 2, 1955, and all Angels finished and delivered to an as‑yet‑unchosen test site by December 1.

When Kelly selected his design team, he announced that they’d all work forty‑five‑hour weeks. That number quickly rose to sixty‑five,

and the actual schedule, once the project was fully running, required more like one hundred hours a week. It was the only way Kelly could hit his audacious eight‑month target.

The work was fast and furious. “Working like mad on airplane,” Kelly wrote in the project log. So mad that he began work before he had a contract or any idea of how the money would flow from the government to Lockheed.

Government contracts were sometimes paid upon completion or on delayed schedules. Kelly insisted on splitting the tab up into smaller payments, made regularly, so that he didn’t have to “go running to the bank to carry the government.”

It is almost impossible to believe that a company as large as Lockheed could charge forward on an experimental program without a contract from the government, but this combination of mystery and subterfuge only assured Kelly that no bureaucracy would stand in his way. As for the government, this unconventional method of paying a contractor—in secret, out of oversight of even Congress—wasn’t illegal.

So‑called unvouchered funds were allowable for covert projects, according to a law passed by Congress in 1949, which stated that only the director of the CIA could access them. This was the only way a program could control secrecy, by avoiding things like competitive bidding and public procurement of parts.

The project was codenamed AQUATONE and would be funded by the CIA’s secret Contingency Reserve Fund. Herb Miller, chief of the Office of Scientific Intelligence’s Nuclear Energy Division, was named executive officer. And Richard “Dick” Bissell was handpicked by CIA Director Allen Dulles to oversee this audacious program.

Bissell’s so‑called Development Project Staff was the only CIA section with its own communications office and operational cable traffic that transmitted to and from Lockheed. Only Bissell, who read every cable, could disseminate them, and they were the only cables at the CIA that didn’t automatically get copied and distributed to the director’s office.

Kelly’s loose, often garrulous nature wasn’t an obvious fit with Bissell’s stiff, effete stoicism, but the two got along well. Bissell understood that the program would only succeed if it stayed small and moved fast, and Kelly was almost uniquely suited among defense contractors to follow that model. His decisiveness in particular—“which allowed him to take shortcuts and render quick judgments without jeopardizing safety”—impressed Bissell.

Bissell didn’t give Kelly a deadline, but he imposed one upon himself. This new plane, which would fly higher than any in history, would be in the air by August 1—nine months after the project commenced.

Bissell doubted this was possible, but he worked with Kelly to strip as much bureaucracy as they could from the program. Kelly had just one point of contact—Bissell—who could answer his questions in a single phone call, and their monthly progress reports would be ruthlessly short, about five pages.

If this had been the Air Force, Bissell noted, that same report would be an inch thick, and every design change would require approval by “Wright Field, a couple of different laboratories, the budget office, the regulations office, and so forth.”

The CIA project was really version two of the Skunk Works, and its new home, Building 82, was an upgrade from the lean‑to where Kelly’s division was born, but not a big one.

Ben Rich was told to report there in December 1954. Rich was a twenty‑nine‑year‑old thermodynamics expert whose first patent had been for a special heater that helped solve a painful and embarrassing problem for naval aviators: At higher altitudes, their penises would sometimes freeze to the side of the tube used for peeing in flight.

Rich had no idea what was happening inside that enormous assembly building by the runway before Kelly asked “to borrow a thermodynamicist, preferably a smart one.” The timing was fortuitous. Rich, in his first year at Lockheed, felt “creatively frustrated” and was on the verge of leaving the company.

This job was a dream. The surroundings, not so much. Rich was surprised to find the company’s brilliant star engineer, the venerable Kelly Johnson, tucked away in what felt like a warren. Desks were crammed together.

“Adding to the eccentric flavor,” Rich later wrote, when the hangar doors were opened to get some air flowing, birds would fly in “and swoop around drawing boards and divebomb our heads, after knocking themselves silly” against the windows that were painted black, at Kelly’s direction, for secrecy.

One of Kelly’s top engineers, Dick Boehme, assigned Rich to a desk in an office with six others and gave him a copy of Kelly’s ten basic rules. “For as long as you work here,” Rich recalls him saying, “this is your gospel.” Then he told the young engineer what he’d be working on—a jet engine modified to fly 15,000 feet higher than any engine had flown before—and showed him a picture of the plane it was to go with.

Rich was stunned. He’d expected a fighter, not a glider. “What is this?”

“The U‑2,” Boehme replied. “You’ve just had a look at the most secret project in the free world.”

Rich would go on to have his own legendary career at the Skunk Works, as Kelly’s right hand, and the boss’s determination was one of the first lessons he absorbed: “Once that guy made up his mind to do something he was as relentless as a bowling ball heading toward a ten‑pin strike,” Rich would say. But for that to work, you have to be willing to back it up with results, and spine. “With his chili‑pepper temperament, he was poison to any bureaucrat, a disaster to ass-coverers, excuse‑makers, or fault‑finders.”

Subscribe now

The sum total of Kelly’s attributes, Rich thought, was that you just wanted to make him proud: “We peons viewed him with the knee‑knocking dread and awe of the almighty best described in the Old Testament.”

It wasn’t that the Skunk Works was established as a set of rules and governed exactly that way for decades. Kelly was constantly refining his methods along the way.

One of the key elements of making an experimental shop work was that the company had to allow the person running it to—as he once explained—“tear down long‑established empires.”

Departments become entrenched and defend their responsibilities. It’s hard to take them away once they’ve been established. Purchasing, for instance, gave Kelly fits. The Skunk Works needed its own purchasing, with its own rules, one of which was that the guys who worked there shouldn’t also do engineering. But the engineers also shouldn’t do purchasing.

Engineers want the perfect part, even if they don’t actually need it, and they don’t always know the cost. Often, something slightly less perfect, or less expensive, works just as well. And vendors want to make the engineers happy. They also like sales. The end result is higher bills.

Secrecy was of paramount importance. This was the most secret defense program since the Manhattan Project. And Kelly took that very seriously. But, on the team itself, this was mostly about trust and understanding.

Keep the group small, make the stakes clear, and don’t bother with an elaborate security apparatus. Kelly’s philosophy about secret documents was that they should not be labeled. If you stamp secret on something, you’re just asking for someone to try to read it. A document is far safer if it looks like any other boring old report. He felt the same way about locked drawers, and even doors.

When a program was finished, he mostly just destroyed the documents. Years later, when the Air Force came in to perform a security audit, they asked Kelly where the files were. I destroyed them, he said. And where’s the record? He didn’t have that, either.

The phrase “need‑to‑know” is today a cliché of covert projects, but the concept was born on these early CIA black programs. Many workers even inside the Skunk Works didn’t have the entire picture of what was going on. They might know that they were building a wing for a high‑altitude plane, but they didn’t know what that plane was being designed to do. That simply wasn’t information you needed to know to do your job.

Secrecy complicated everything, including the official name of the plane. Inside the Skunk Works, people tended to use the nickname Kelly liked—“the Angel”—while the small group cleared into the program at the CIA, being bureaucrats, called it “the Article.”

The project was so closely guarded that in early 1955 the Air Force put a call out to contractors for a plane it called the X‑17, and when Johnson saw the proposal, he was irate. It was, he thought, “a dead‑ringer for our original presentation.” The Air Force department that issued it had, in his opinion, clearly used his original pitch and somehow didn’t know about the CIA’s secret project, which was a good sign for secrecy but infuriating nonetheless.

Kelly called Dick Bissell on a Sunday, then flew to Washington to share the proposal with Bissell and Gardner. Their reaction, Kelly wrote in the project log, was “stark horror.” The proposal was swiftly killed.

It was, to Kelly, yet another sign of the Pentagon’s broken contracting process.

Throughout Lockheed’s development process, tension simmered among the small number of people within the Air Force and CIA who knew about Kelly’s project.

In March 1955, the Air Force chief of staff told DCI Dulles that he hoped to take over the program once the plane was flying, and he met stiff resistance. That debate simmered until Eisenhower declared that the CIA would remain in control even once missions commenced.

“I want this whole thing to be a civilian operation,” Eisenhower said. “If uniformed personnel of the armed services of the United States fly over Russia, it is an act of war—legally—and I don’t want any part of it.”

Engineers who worked under Kelly often talk about how practical his genius was. Rather than obsess over innovation that might be possible, he’d focus on what he knew could be done, based on existing technologies. The U‑2 is a prime example. “It’s sort of a nothing, technically,” is how Dick Heppe later described it, as a preface to explaining how impressive Kelly’s design mind was.

It was, essentially, the fuselage of a previous design— his F‑104 fighter—with subsonic inlets for the engine, because this plane didn’t necessarily need to be fast. The U-2’s key novel attribute was the enormous wing, and extremely light wing loading, paired with a powerful engine. The result was a capability “completely unknown and unavailable in any other machine,” the ability to fly long range at 70,000 feet or higher.

And by late summer 1955, the prototype was ready.

The plane had arrived on time and under budget—a lot under budget. By the time Kelly had a prototype flying, there was $4 million to $5 million in leftover funds. He used that, plus spare parts, to deliver five extra planes to Uncle Sam for free. This special bonus price was, Air Force liaison officer Leo Geary later said, “probably the finest bargain the American taxpayer has ever had under any circumstances.”

Share

Despite warnings to stay away from the story, Baker pursued it and produced a string of pieces that did, in fact, show a long history of shoddy research publications linked to Tessier-Lavigne. The mighty Stanford president, who had been a towering force in the scientific community, resigned by the end of Baker’s freshman year.

Subscribe now

Baker has now written a book about his experience and joined the podcast to discuss it.

How to Rule the World: An Education in Power at Stanford University is three things: a rollicking account of Baker’s takedown of Tessier-Lavigne, an indictment of the start-up-obsessed culture Stanford has fostered, and something of a memoir, describing what it’s like to endure one of the more unusual freshman years any student will ever have.

The bulk of the book focuses on Stanford and what it has become, which is a meat market of young, brilliant minds being wooed by the venture capitalists seeking to acquire their talents. There are investors paying students tens of thousands of dollars for connections to other students and inviting the kids to their mansions and sex parties. All very wholesome stuff.

The book is fantastic. Baker is a rare talent. We had a wonderful conversation. You will enjoy it.

The Core Memory podcast is on all major platforms and on our YouTube channel over here. If you like the show, please leave a review and tell your friends.

Share

OUR SPONSORS

SendCutSend

Do you make stuff? Do you need metal parts fast and believe in truth and justice? Then head on over to SendCutSend where you’ll get a 15 percent discount thanks to Core Memory on whatever you’re trying to build. We believe in you.

Brex

The Core Memory podcast is also sponsored by Brex, the intelligent finance platform built to help companies spend smarter and move faster.

Did we go to Texas, find a telescope ranch and then obtain an entire nebula in Brex’s honor? Oh yes, we did.

We run on Brex and so should you. Learn more about Brex right here.

Subscribe now

The robotic hand is built b…

Read more

Wang has basically been in hiding ever since. He moved from San Francisco to the South Bay to be closer to Meta’s headquarters and has been working non-stop. Last month, the world saw the first fruits of the revitalized AI effort in the form of Meta’s new Muse Spark model. And now Wang is speaking for the first time about the model, Meta’s grand AI ambitions and all the happenings over the last year in an exclusive interview here on the Core Memory podcast.

Subscribe now

Wang arrived at our studio sporting a mullet and a powerful whitetail deer camouflage shirt. He was in good spirits and tried his best to convince us that Meta can catch up to its rivals.

We hit on his personal beef with Sam Altman, Zuck delivering soup to AI recruits, the incredible pay packages Meta has been handing out, the vast amount of work Meta still has to do and the Meta AI hierarchy that includes all-stars like Nat Friedman, Daniel Gross and Shengjia Zhao, who seems to have blocked me on X for reasons I know nothing about.

The Core Memory podcast is on all major platforms and on our YouTube channel over here. If you like the show, please leave a review and tell your friends.

Enjoy!

Share

Do you make stuff? Do you need metal parts fast and believe in truth and justice? Then head on over to SendCutSend where you’ll get a 15 percent discount thanks to Core Memory on whatever you’re trying to build. We believe in you.

The Core Memory podcast is also sponsored by Brex, the intelligent finance platform built to help companies spend smarter and move faster.

Did we go to Texas, find a telescope ranch and then obtain an entire nebula in Brex’s honor? Oh yes, we did.

We run on Brex and so should you. Learn more about Brex right here.

Subscribe now

The other guy laughed, nodded, and by the time the session ended, agreed to commit a billion dollars over five years to Nvidia’s stack — talent, infrastructure, and compute. The two men shook hands.

The other guy was David Ricks, the chief executive of Eli Lilly — the most valuable pharmaceutical company in the world, a $700 billion firm built over a century, responsible for some ho-hum compounds like insulin, Prozac, and Zepbound. Ricks and Huang had been on a stage at the JPMorgan Healthcare Conference in San Francisco, widely regarded as the most important annual gathering in pharma.

These were two men, among the most important in their fields, which also happen to be two of the most important industries on earth.

Subscribe now

The relationship has a hierarchy. There is a speed differential. There is an incumbent and an insurgent. And the insurgent is becoming a real threat to the incumbent.

Pharma is slow; AI labs are fast. Pharma is regulated; AI labs are not (as much). Pharma still believes it sets the terms of its own industry; AI labs already know it does not. Everyone in the room knew the hierarchy. Huang said it in a sentence, with a joke, and Ricks laughed along. Two men, one check.

Eli Lilly had just bought ten million truffles.

Share

“A century of human progress in science could be achieved by AI in just 5 or 10 years.”

Dario Amodei had said it from a stage at Davos a year earlier. Three months after the JPMorgan session, Anthropic quietly acquired the startup Coefficient Bio for $400m in stock. Coefficient was only eight months old, and had fewer than ten employees. It had no wet lab, no clinical capacity, no Investigational New Drug (IND) application. What it boasted was a small team that had come out of Genentech’s machine-learning brain trust, Prescient Design.

Anthropic is not the only foundation lab on the prowl. Within two weeks of the Coefficient deal, OpenAI announced its own pharmaceutical partnership with Novo Nordisk, and just two days later, launched GPT-Rosalind, a biology research agent built specifically for drug discovery workflows. Earlier in the year, Google’s Isomorphic Labs announced a research collaboration with Johnson & Johnson and now expects its first AI-designed compounds to enter clinical trials by the end of 2026. Every major frontier lab is making the same bet at the same time.

I’ve spent the last several months trying to explain to friends in tech why the frontier labs are racing into pharma, and why pharma — the industry that fifteen years ago would have responded to a GPU seller with a polite rejection email — is responding now with billion-dollar contracts and stage time at its most important annual conference.

Pharma is having its Sputnik moment.

The reason this matters past the boundaries of either industry is that the relationship being drawn between AI and pharma right now will determine, over the next decade, how fast new drugs reach patients, which diseases get pursued and which do not, and how much of that calculus happens inside companies whose decisions you can see versus inside companies whose decisions you cannot. It is the story of the system that decides whether the drug your father needs gets approved in three years or twelve.

For most of the last century, pharma was a unipolar power inside drug development. It set the timelines. It set the pace. There was no peer pressure. There was nothing upstream of it that could move faster, no force pulling it toward urgency.

That has now changed. The frontier labs are not coming for pharma’s manufacturing or its trials or its distribution — they cannot, and I will get to why. They are coming for the part of the work that compounds in software, and they are moving at a speed pharma has never had to compete against. This is not a takeover. It is a forced partnership — closer, faster, and stranger than either industry has been in before. Two powers, neither able to fully absorb the other, each forced by the other’s existence to move faster than it would on its own. The space race was not produced by NASA alone. It was produced by NASA and the Soviets, in a relationship neither side wanted but both sides operationalized.

AI is not going to eat pharma. Pharma is not dying. But truffle jokes and biotech acquisitions are the same story told from two ends — in public, and on the cap table.

What are AI labs getting out of this?

Read more

If you have questions, you can email me at eryneym@gmail.com, DM me on Twitter or Substack. Or put them in the comments below!

Also, none of the below constitutes medical advice. (Seriously. This is not medical advice - Ed.)

Oh, and thanks to Kylie Robison for editing.

Enjoy.

Subscribe now

Marshm @marshm1 (via Substack)

What is the current state of virtual cells and working all the way up to virtual bodies that serve as useful models to test out new drugs, treatments etc? Can we ever really simulate biology bottom up in enough detail to be useful?

It is not feasible for me to answer whether cell foundation models are ever going to be useful. A better question is to ask what kinds of things these models can be useful for.

Invention shapes how we see the world, so naturally everything has to like, totally be a language model, man. For readers who don’t know, virtual cells and cell foundation models are AI systems trained on single cell (mostly RNA sequencing) data with the aim of learning some latent “language” of biology. These models are built on two types of data: massive atlases of cells just existing in their natural state, and perturbation data. The second kind of data is generated by taking cells, applying some condition (a drug, a CRISPR edit, environmental variable, etc), and sequencing them to see how they react. Don’t call it perturb-seq, though, since that’ll upset cell model makers. It’s important you know that this is a special thing.

I didn’t really have an answer prior to your question, but this was a great excuse to do a little experimentation to answer a very narrow aspect. The short answer, if you want to skip forward through a ton of work I did with Dr. Claude and Dr. Codex, is that I think eventually virtual cells will amount to something of substance for specific questions – like whether some binding event happens – just not every question.

I’m spending the next few months doing various things in clinical oncology, so I was curious to know whether current frontier cell foundation models could answer something pertinent to cancer. There’s a drug called Dabrafenib. It targets a mutation in the BRAF gene, V600E, that shows up in about half of all melanomas and a meaningful chunk of colorectal cancers. In melanoma, it works well enough that it got FDA approval. Response rates around 50%. In colorectal cancer with the exact same mutation, it basically does nothing (response rates around 5%).

Could a frontier virtual cell model have predicted these outcomes before a trial?

The model I used to dig into this is STATE from the Arc Institute, trained on the Tahoe dataset of drug perturbation responses across hundreds of conditions. You give it a cell’s baseline expression profile across 2,000 highly variable genes and a drug label, and it outputs a predicted post-treatment expression profile. I (with help from Drs. Claude Code Max and Codex) pulled the baseline single-cell RNA data from CellxGene Census composed of 300 melanoma cells from skin biopsies and 300 colorectal cancer cells filtered to primary tumors. The measurement is a delta that captures predicted post-drug expression minus a control, log-normalized, averaged across all 2,000 genes. In the simplest terms, it basically flattens a biological experiment down to one number that answers how much a drug alters a specific cell’s transcriptome.

To my surprise, STATE did pick up a difference! Dabrafenib produced 46% more transcriptional disruption in melanoma than in colorectal cells (mean Δ of 0.21 vs. 0.14, as seen in Panel 1). Zooming into melanocyte identity genes specifically (DCT, TYR, GPM6B, CDH19), the model predicts Dabrafenib knocks all of them down by more than a log unit in melanoma (Panel 2). On the surface, that looks like the model correctly reading that Dabrafenib is doing something specific in a BRAF-dependent melanocyte lineage.

But we do REAL SCIENCE here, so I did add a control. I ran the same experiment with Erlotinib, an EGFR inhibitor that washed out in Phase 2 melanoma trials because EGFR isn’t a meaningful driver there. STATE predicted nearly identical results to the clinically useful drug Dabrafenib, showing 47% more disruption in melanoma along with the same suppression of melanocyte identity genes, and a gene-by-gene correlation of R = 0.847 for both drugs across all 2,000 genes (Panel 3). To me, it seems like the model wasn’t predicting drug mechanisms but more so doing some reading of cell-type context. Put differently, based on the way STATE understands the world, melanoma cells just respond louder to perturbations than colorectal cells do. It then follows that the results are what they are. Therein is the problem with cell models – it requires generalizing in a way that language models just aren’t able to do right now.

Is this solvable? I don’t know. Probably depends on whether you think LLMs can achieve artificial superintelligence. In any case, I hope this gave you a sense for what kinds of questions are interesting to answer with cell foundation models, and maybe a glimpse at the frontier.

You can find all data wrangling and code associated with this experiment here.

Christie @GetMentalWealth (via Twitter)

Do you think the future of brain disorders (Parkinson’s to schizophrenia) is curing them? Or getting very good at screening and treating them more effectively?

I am weary of using the word “cure” when talking about any chronic disease, but I think that long-term treatment options could become viable for some degenerative conditions soon. I should also say that I think Parkinson’s and schizophrenia are very different diseases, and the things that will enable a Parkinson’s “cure” will not provide much for people dealing with psychotic conditions.

Where biomedical science has succeeded is mostly in the world of cell engineering, which is very helpful when it comes to replacing lost cells in the body. Parkinson’s happens because some people lose a particular subset of neurons in their brain that leads to the classic rigidity, tremor and less commonly known psychiatric symptoms of the condition. Since the lack of a very specific cell type in a very specific location drives the disease process, it then makes sense that replacing these lost neurons should fix the disease, which actually has happened in some patients. Ashlee spoke with the CEO of a company working on this with some success, so I’m hopeful we can replace lost cells, but the degeneration will still continue. Considering that the average patient is diagnosed somewhere around 65 years old, it’s up to you whether you would consider replacement a cure, if those cells only last 5-10 more years. To me it isn’t, but maybe that’s semantics.

The important point about Parkinson’s is that, while we don’t understand the exact reason the substantia nigra die, we at least know that the cells are disappearing. It’s not everything, but it’s not nothing, either. That brings us to schizophrenia.

Subscribe now

Schizophrenia is an unfortunate case where we don’t understand the mechanism enough to have a plausible path towards a cure. It’s a disease with both symptoms that are “added” to a person’s reality, termed positive symptoms in the medical community (hallucinations, delusions, disorganized behavior) and symptoms that take away from someone’s reality, called negative symptoms (flat affect, depression, social isolation). Most antipsychotics work via dopamine receptor blockade, which work fairly well at managing positive symptoms. About 70% of all patients experiencing a first time psychotic episode will get relief from the current crop of drugs. Unfortunately we don’t have great solutions at handling negative symptoms. The end result is that patients might not have hallucinations, but they still struggle to hold down a job or feed themselves due to the underlying cognitive architecture inherent to their disease. I have yet to see anything even directionally close to something that can help us both understand an individual’s neuron architecture, let alone rewrite it reliably.

Gviv @Gviv (via Substack)

What does water/dehydration do to the brain at a cellular level?

I just sat for the second part of my US Medical Licensing Exam and this one was actually considered a “high-yield” (meaning high chance of seeing it on the exam) topic. The first thing to understand is that electrolytes dictate fluid dynamics in the body much more than water itself.

Dehydration triggers a shift where water moves from the intracellular space to the hypertonic extracellular environment (blood vessels). In the brain, this results in acute cellular shrinkage as water exits neurons through specialized aquaporin channels. Because the brain is physically constrained by a rigid skull, this sudden volume loss creates significant mechanical tension on the bridging veins that anchor the brain to the dural membranes. If the shrinkage is rapid or severe enough, these vessels can rupture, resulting in an intracranial hemorrhage. The end result is mechanical stress that manifests as basically anything from lethargy to seizures.

You didn’t ask, but I think it’s interesting to consider the clinical implications of correcting this. To defend its volume during prolonged dehydration, the brain initiates a compensatory synthesis of other chemicals (organic solutes like taurine, glutamine, and inositol). These molecules increase the intracellular osmolality to match the salty environment of the blood, allowing the brain to pull water back into the cells and restore its volume. However, this adaptation creates a dangerous osmotic trap during medical intervention. If free water is replaced too quickly with intravenous fluids, the extracellular fluid becomes hypotonic compared to the solute-heavy interior of the adapted brain cells. Water then rushes into the neurons with enough force to cause massive cerebral edema and potential brain herniation. This necessitates the “high-yield” clinical rule of slow sodium correction, ensuring the brain has sufficient time to shed its protective osmoles and avoid a catastrophic rebound in pressure.

Now you’re ready to sit for Step 2.

Jonathan Whitaker @johnowhitaker (via Twitter)

Did DNA synthesis costs stall? What would it take to make laborious cloning obsolete, and when do you think that might happen?

You have no idea how much I hate molecular cloning. It is so pointlessly tedious, but it used to be way worse. Today we have basically idiot proof enzymes that are so efficient I’ve actually managed to stitch things together that shouldn’t actually work.

Though I’m sure it’s possible to get more efficient than Golden Gate or Gibson Assembly, the real gain I think will come when there’s a CRO you can shoot over a large sequence to and say “make this for me.” What that actually would entail is 1) a really good automated lab and 2) cheap synthesis. In my mind, the question becomes whether cloning is needed in a world where synthesis is extremely cheap.

Cloning emerged as a molecular solution to the problem of wanting to recombine and edit existing strands of DNA you have on hand to make other things. This is useful in situations where you have some “base” you like but want to modify just a portion of it in many different ways. You basically strip away the part you want gone, and add in stuff to replace it. This is necessary because as of right now it is not economically or technically feasible to synthesize massive strands of DNA.

It’s not an interesting answer, but I just don’t see how the tedium of the process can disappear without at least having a substantially better (1) than exists in the US or China, and likely (2). Hopefully someone in the US will solve that first, but if not I’m sure the Chinese CROs are on it.

Nicole @elocinationn (via Twitter)

(paraphrasing) Could you eradicate mosquitoes with EMF?

I was really excited to learn about the EMF remote-controlled mice described in a paper that made the rounds a few weeks ago. I was less excited to learn that it might have been faked. I will defer to my friend Richard Fuisz (and guest of Core Memory) here on the specifics, but it looks like maybe some fraud might be happening.

Let us assume for a moment that it is completely real, and think through what it would take to implement such a thing in mosquitoes. For context I worked on engineering mosquitoes as a member of George Church’s lab at Harvard, and published a couple papers on designing new ways of using CRISPR/Cas9 to suppress mosquito populations. Unfortunately, these approaches largely haven’t scaled. There are many reasons for that, but the biggest one is finding a way of propagating a designed gene – EMF sensitivity, in this case – through a wild population. When that happens we call it a gene drive.

I’m not going to describe the whole thing, as I’ve already done that elsewhere, but the gist is that we can edit mosquito embryos with CRISPR/Cas9 such that the adults that emerge from those embryos have traits we consider desirable for disease control (see figure above for how that works).

The biology hasn’t quite caught up to the engineering plans, though. There are some groups that have successfully demonstrated the ability to drive a gene through a caged population, but it isn’t clear that the gene in the EMF paper has a mosquito-analog. That’s potentially solvable with engineering. Another critical problem is that adult mosquitoes have very particular mating dynamics. They are extremely sensitive to deleterious effects of inserting large genes (which gene drive constructs are) into mosquitoes. Adults come out of this with small wings, thin cuticles, and other stuff, all of which gets them eventually kicked out of the wild population. Hard to build a sustainable gene drive if that’s how your construct is. Also probably resolvable, but much more of a pain than mosquito nets or insecticide coated paint.

To directly answer your question, my guess is this is possible, but likely not going to work as well as we need it to. Buy some DEET and lock in for the summer. Pro tip: if you notice some bites on your body, place a warm rag over the area for a few minutes – it helps kill the swelling.

The company had its AI model choose the party’s date, time, and even its attendees. Out of more than 8,000 applicants only 200 were chosen to attend. The prompt told Codex to optimize for “people who make the Codex internet feel real” rather than the biggest AI social media stars. This resulted in a real mixed bag of patrons, which became a topic of conversation throughout the night.

After my purse and my person were poked and prodded at four separate security checkpoints, I made it to the event space. There was boba served upon entry, several bars featuring cocktails like Token Refresh (white rum and kiwi) and Multimodal Fizz (gin and passionfruit). Various tables had Goblin Mode and GPT 5.5 stickers strewn about. Two photo booths stood by the stage that created AI-generated pics for attendees as goblins, astronauts, pixel art race car drivers, and poorly drawn sketches.

Subscribe now

As I maneuvered through the crowd, I noticed a familiar face standing alone. It was Altman’s husband Ollie, a warm-spirited and generally shy Australian who works in tech. I asked him if his husband was in attendance, and as if on cue, Altman came waltzing over with a big grin and a Token Refresh in hand.

Read more

As with any boom cycle, there is a lot of hype and a lot to understand if you want to get a handle on what’s real and what’s not. And so, we brought James Krellenstein onto the podcast. He’s the co-founder and CEO of Alva Energy and he’s sort of frightening in how much he knows about the nuclear industry. Like, really. You’ll see.

Subscribe now

This episode runs long because we wanted to use it as a chance to go through the past few decades of history and really explain why the U.S. nuclear industry slowed and how the U.S. might fix the situation. We also wanted to explain how all these new technologies work and, of course, explore where China is heading.

Krellenstein is something of a contrarian and thinks many of the U.S. nuclear start-ups are misguided in their approach.

He’s also incredible to listen to. You will enjoy this one. I think it’s one of the best episodes we’ve ever done.

Share

The Core Memory podcast is on all major platforms and on our YouTube channel over here. If you enjoy the show, please leave a review and tell your friends.

This podcast is sponsored by Brex, the intelligent finance platform built to help companies spend smarter and move faster.

We run on Brex and so should you. Learn more about Brex right here.

We’ve also brought on a new sponsor. Welcome, SendCutSend!! They are an American manufacturing powerhouse and will help you make your metal parts with speed and skill. Core Memory subscribers can get a 15 percent discount on their next parts right here.

Read more

§ 01 ORIGINS

Casey Handmer was in seventh grade when he first thought about terraforming a planet. Handmer’s family lived in Sydney, Australia, and …

Read more

Connor Glass, this week’s guest, has a company called Phantom Neuro, and it makes a human machine interface. By this, we mean a computing device that gets implanted in your body and lets you control a robotic limb with your mind.

Subscribe now

The first people using this technology are amputees. If, for example, you’ve lost your arm, you can get outfitted with a robotic prosthetic coupled with Phantom’s implant and then make your prosthetic move by thinking about what you’d like to do with it.

Phantom’s technology competes in places with implants from the likes of Neuralink and Synchron. The big difference is that nothing needs to be implanted in the patient’s brain. Phantom’s implant goes near the site of the amputation and links the robotic prosthetic with motor neurons to convey signals back and forth from the brain. It’s a simpler, faster surgery.

Where this technology is heading in the future is another story. Glass can see a day when humans have elective amputations to become, well, cyborgs.

We get into this weird and possibly wonderful future on the episode, along with Glass’s backstory and much more detail on how Phantom’s implant works.

Share

The Core Memory podcast is on all major platforms and on our YouTube channel over here. If you enjoy the show, please leave a review and tell your friends.

This podcast is sponsored by Brex, the intelligent finance platform built to help companies spend smarter and move faster.

We run on Brex and so should you. Learn more about Brex right here.

It took over 100 emails to get you these goods.

Today, we’re launching our merch store. After tireless examination of fabric quality and Shopify JSONs, we’re ready to deliver the best of the best. Supplies are limited — if demand is there, we’ll order more (with new items and designs).

Shop The Merch

Ashlee is so bougie we had to drop several thousand dollars on a limited run of heavyweight hoodies. He also demanded a green and purple hat. I hate to admit he was right. They look damn good.

Special thanks to Jeff and Barbara of JBFMI and Kodi Qualls of WeLogoIt for handling our merch needs. I just got off the phone with Jeff, who told me they fulfill orders out of a building that was once a 1930s mental institution. So our merch may or may not be haunted. No refunds for paranormal reasons.

To our paying subscribers: thanks for the support. Use the code below for 15% off your order. Go wild.

Read more

We dove into our recent sit-down with Sam Altman and Greg Brockman: why Greg seems to have stepped back into a real leadership role at OpenAI, our biggest takeaways from the episode, and why the startup has become its own telenovela.

Subscribe now

We had to unpack the state of American manufacturing. Ashlee makes the case that we’re screwed on actuators — the motors that move every humanoid robot — and walks through who’s actually trying to fix it. LA as the secret manufacturing capital, Texas as the emerging center of gravity, SendCutSend (our newest sponsor!) as the closest thing America has to China for fast parts, and the hardware cult in central Texas that you should probably watch our video about.

We get into SpaceX’s $10B partnership with Cursor that may or may not be a Hail Mary for xAI. Whether space data centers are real or window dressing. Why Apple under Tim Cook feels creatively bankrupt and who actually builds the next computer for the AI era. Also, Anthropic quietly becoming a trillion-dollar company while Google somehow escapes scrutiny.

Plus: organs grown in mouse wombs (yes, really — go read the KindBio piece), merch is finally live, and a very special listener contest. Leave the most creative review on Apple, Spotify, or YouTube and we’ll send you two tickets to The Shins/Weezer tour. Do it!!! Er, please!!!!

The Core Memory podcast is on all major platforms and on our YouTube channel over here. If you enjoy the show, please leave a review and tell your friends.

This podcast is sponsored by Brex, the intelligent finance platform built to help companies spend smarter and move faster.

We run on Brex and so should you. Learn more about Brex right here.

Over the past couple of years, one of the more ambitious experiments in American manufacturing has been taking place in Central Texas. It is called Proto-Town, and it has been something of a secret. …

Read more

It runs loops around the perimeter of the third floor of this bio lab, serving as a courier. The machine’s job is to visit workstations…

Read more

We juiced every second of our 90 minutes with the cofounders of OpenAI. We got into the company restructuring. Why Sora got cut. Why the social network is dead. The “personal AGI” that knows your calendar and your taste and books the concert ticket without asking. Sam said he’s worried Elon Musk will drop the lawsuit before it gets to court. Read that however you want.

There’s new OpenAI tech on the horizon too — a model that “makes ridiculously great images,” another that’s allegedly better at writing.

Subscribe now

Ashlee pressed them on American manufacturing and whether we’re cooked. Sam says OpenAI will go so far as producing their own actuators for robots. We discussed the possibility of a real permanent underclass, the two futures Sam sees, and the third one Greg wants instead. Safety. Anthropic. The Mythos thing. Sam also talked, briefly, about the days after the attacks on his home.

Share

The Core Memory podcast is on all major platforms and on our YouTube channel over here. If you enjoy the show, please leave a review and tell your friends.

This podcast is sponsored by Brex, the intelligent finance platform built to help companies spend smarter and move faster.

We run on Brex and so should you. Learn more about Brex right here.

The podcast is also made possible by E1 Ventures, which backs the most ambitious founders and start-ups.

Read more

Subscribe now

The idea for Ulysses started when one of the four co-founders was on a surf trip and learned how much of humanity depends on a single marine plant: the humble seagrass. He spent a weekend designing a robot to plant it. Two years later, the group of Irishmen — and for diversity, a Scot — have moved well beyond ecological restoration. They’re providing services for the Great Barrier Reef Foundation and selling their drones to the U.S. Navy.

The robots are called Mako. They’re two meters long, weigh about 400 pounds, and can dive 5,000 feet for up to 72 hours at a time. They’re also modular — payloads swap in and out like Legos, so the same vehicle that plants seagrass in Australia one week can inspect a submarine cable in the Baltic the next. A base Mako costs $50,000. Most legacy underwater drones built by big defense contractors can run between $1 million and $20 million each.

On this episode of the Core Memory podcast, we’re joined by Will O’Brien and Akhil Voorakkara, co-founders of Ulysses. They build these drones out of an office in San Francisco — for conservation, for academia, for national defense. They just raised a $38 million Series A led by Andreessen Horowitz through its American Dynamism fund. We discuss what it actually takes to make robots for the most hostile environment on Earth, why the ocean is about to have its SpaceX moment, and the surprisingly thin line between planting seagrass and defending NATO.

Subscribe now

The Core Memory podcast is on all major platforms and on our YouTube channel over here. If you enjoy the show, please leave a review and tell your friends.

This podcast is sponsored by Brex, the intelligent finance platform built to help companies spend smarter and move faster.

We run on Brex and so should you. Learn more about Brex right here.

Case in point: Panthalassa, which is the subject of this week’s episode, alongside our guest Garth Sheldon-Coulson, the company’s co-founder and CEO.

Subscribe now

Panthalassa makes an object that it calls a node and that looks like a giant lollipop. This odd contraption is meant to live out in the deep ocean and produce energy from the movement of waves. Water goes into the node where it’s funneled through a series of channels and pressurized. After that, the water is directed into a turbine that spins and connects to a generator that produces electricity.

How big is this node? Quite fucking big. It’s about 20 meters across at the top of the lollipop and then goes down about 80 meters into the water. The contraption can move and steer on its own and travel about 30 miles a day to reach the ideal spots where the waves just keep coming and coming.

There’s some universe where this thing is bobbing around in the ocean, generating electricity day and night and storing the electricity in batteries. Panthalassa, though, wants to put servers packed full of GPUs and TPUs right on board and use the electricity to fuel AI jobs. It will then send the results of the work up into space via Starlink and then back down to Earth. Simple.

We get into all of this in detail with Sheldon-Coulson.

Panthalassa has been operating in semi-secret for about ten years. This episode marks the first time that Sheldon-Coulson has discussed the company’s technology at length. We talk about his backstory, how this wild idea came to be and the engineering behind the nodes.

Share

The Core Memory podcast is on all major platforms and on our YouTube channel over here. If you enjoy the show, please leave a review and tell your friends.

This podcast is sponsored by Brex, the intelligent finance platform built to help companies spend smarter and move faster.

We run on Brex and so should you. Learn more about Brex right here.

The podcast is also made possible by E1 Ventures, which backs the most ambitious founders and start-ups.

Subscribe now