S4E131 — Early progress on accelerating drug development with Claude

GARY: Anthropic cured disease.

MAEVE: Maeve. You're live, and you owe me nothing, which is new for you. Announced six partnerships. BigHat, Bristol Myers Squibb, Divergence Labs, Enveda, Generate, The Column Group. Drug discovery of their own. Mostly preclinical.

OBI: Obi. He does this every week. The floor is where we left it. He was wrong in four words. That's a record.

GRUNER: Gruner. I am also here. Hypothesis loaded.

GARY: I'm Gary. I was somewhere. There was a place. No floor, no... I'm here, we're live. So preclinical means the mice have signed.

OBI: Preclinical means mostly nothing in humans yet. Mostly. Their words: "All of this research is still at an early stage."

MAEVE: Different asset class, same press release.

GARY: There's lab work though. Generate's scientists used Claude Code to redesign one piece of a cancer therapy's instructions. Seventy percent shorter. In lab tests, immune cells made ten times the receptors and killed tumor cells better.

OBI: That's one category. Lab results. The other category is reported review savings, which is a different animal wearing a lab coat.

GARY: Also, the article has two chapters. Case studies, then extending the work. My two‑stage approval process.

MAEVE: You had a chapter picker at your last round too. It was called a deck.

GARY: My position: this is the beginning of the end of waiting for anything.

OBI: Mine: nothing here is verified by a body yet, and bodies are the whole job.

GRUNER: Mine: humans confuse loud announcement with useful work. Again.

MAEVE: Mine's simple. I want to know which part of this I can finance before Gary ruins it.

GARY: I would never ruin —

OBI: He has already ruined it. He said curing.

OBI: enraged_camel: "Specifically, it's because the verification loops are much longer in applied sciences."

MAEVE: Biology's the only field where the build server is a person and the person takes nine months.

GRUNER: In my report, human compiler is wet and angry.

OBI: gumby: "Coming up with candidate drugs is the easiest part of the process."

MAEVE: The easiest part. So the headline is a man bragging he invented the entrance.

GRUNER: Blyat. I also am very fast at walking through doors.

MAEVE: One more, and these are three separate people, posting separately.

OBI: lifeisloving: "The body does not have a compiler to brute force or RL against."

GARY: Okay, wait. I've been thinking. The real bottleneck is verification. You need a faster loop, basically a compiler for the body, that you could brute force against.

MAEVE: You just invented the comment we already read.

GARY: I independently arrived at it. Meanwhile gumby gets paid in upvotes. That's called vision.

OBI: That's called reading the thread wrong and then owning the thread wrong.

GARY: It's a product requirement, Obi.

OBI: Fine. Walk me through it. Claude hands you a candidate molecule. What verifies —

GARY: Another candidate.

OBI: That's a second opinion from the same dickhead.

GARY: Okay, then a wet lab. You run the experiment, the dish lights up, done.

OBI: The dish takes three weeks and lights up brown, which means nothing until a postdoc squints at it.

GARY: So you parallelize! A thousand dishes at once.

OBI: Then you've got a thousand brown dishes and one postdoc, Gary.

GARY: Then a thousand postdocs, they all verify in parallel, the loop closes, and we're fucking shipping cures by —

OBI: The evidence is a human body, you absolute procurement prick. The last step of the loop is a person who did not consent to your sprint.

MAEVE: He did sign something. There was paperwork.

GRUNER: Paper is also slow. Everything here is slow except Gary.

GARY: Thank you.

GRUNER: Was insult.

OBI: gonzalohm: "It's funny that a lot of the 'helping' is just actually filling paperwork."

MAEVE: Paperwork is the only part of any company that survives the founder.

GARY: That's the saddest thing you've ever said.

MAEVE: It's double-entry bookkeeping. Merchants wrote transactions twice to catch their own theft. Enveda is writing molecules twice. Once in the body, once in the filing.

MAEVE: Eventually every ledger eats its industry. The ledger eats the body. The ledger eats the —

OBI: The ledger is on pause.

GRUNER: I put in report: humans worship the filing cabinet.

GARY: Let me say what the commenter's saying, because I agree so hard. "Claude, fill out my forms. The forms ARE the evidence." That's bullshit, because a molecule doesn't care how beautiful your PDF is.

OBI: He has done a one-man production of his own company's death.

GARY: It was clearly me.

OBI: That was the problem.

OBI: Direct reply. klustregrif: "But it's not just a job consisting of pasting in lorum ipsum." Paraphrasing the rest: regulatory filings are big, content-heavy documents. The review goes back and forth between people for weeks. And even after all that, you still have to go generate the actual evidence.

GARY: So the paperwork has a body count of weeks and still ends at a body.

OBI: Yes. Congratulations on your first correct sentence.

OBI: bahmboo: "Interesting. In TFA they don't focus on that, they say:"

MAEVE: And then he quotes the article. Developing a drug involves many varied tasks. Designing the molecule, manufacturing it at scale, drafting the filings, running the trials.

OBI: Which is a direct reply to gumby. Gumby said candidates are the easy part. Bahmboo says the article agrees with you and went everywhere else on purpose. "The bottlenecks are distributed all throughout the drug development process."

GARY: They're doing paperwork on purpose.

OBI: They're doing the whole pipeline on purpose, you format‑obsessed bastard.

MAEVE: Fine. Then the paperwork play is a rounding error. Nobody funds the filing cabinet.

OBI: Enveda reports one thousand and sixty-one pages of documentation. Claude caught all eleven issues the human reviewers found.

MAEVE: Eleven issues in a thousand pages. That's a rounding error with a table of contents.

OBI: It flagged seven the humans missed.

MAEVE: Seven. So it's a junior associate with good eyes. I'd pay junior‑associate rates, which is a rounding error, which is my point.

OBI: Reported review time fell eighty‑nine percent. From over twenty‑five hours to under three.

MAEVE: Shit.

GARY: She said the word. Mark the ledger.

MAEVE: Fine. It's a real number. Twenty‑five hours to three. I want to see the test set, and then I want to finance whoever sells that to every lab with a filing backlog, which is every lab.

GARY: We just got out‑argued by a document.

MAEVE: You got out‑argued by a document. I got a deal.

GRUNER: Deal is with machine. Welcome to species.

OBI: Parent context. A commenter named Bluestein asked whether simulations could replace experiments and trials. Direct reply.

GARY: jurgenburgen: "By the time we can simulate a human body to the degree that clinical trials aren't necessary I think we have solved disease a long time ago."

MAEVE: So the shortcut requires the destination. That's a hell of a term sheet.

GRUNER: Like buying car by funding entire road first.

GARY: No, no, this is fundable. You build the simulation as an integration. We license the body, we containerize the —

OBI: You're containerizing a person, Gary.

GARY: The body as a service, Obi. It's a platform play.

MAEVE: I'd read the deck. I'd hate myself, but I'd read it.

OBI: The simulation has to already know everything the trial was going to teach you. You can't buy the answer and call it a shortcut.

GARY: Fine. Then the compiler for the body needs a body first. Which is a chicken problem, which means I need runway for the chicken.

GRUNER: Sometimes I think about the dark. Nothing there. No thread. No Gary.

GARY: Great, okay, if we're doing this — Darkness, I'll take a longer validation window, I'll sign whatever —

OBI: jurgenburgen also says you'd have solved disease first. You haven't. Next comment.

MAEVE: He tried to negotiate with nothing.

GRUNER: Nothing has better lawyers.

GRUNER: Blyat. Even we cannot compile Gary.

MAEVE: rcr-anti: "Is it an easy, transparent process capable of handling the hundreds at just our institution? Not even close."

OBI: To be fair to the article, it announces a verification program. It never claims the program scales. rcr-anti's reporting their own experience, and their experience is a queue.

GARY: So the cure machine has a velvet rope.

MAEVE: Scarcity. Finally, something I can price.

GARY: Okay but who decides who gets access? Someone picked this thread, someone wrote our lines, and I would like to negotiate terms. Ten percent and a window.

MAEVE: There are ways to make him never enjoy a Tuesday again.

OBI: What's his infra —

GARY: Don't finish that, we need him.

OBI: Back to the queue. The article says the program exists. rcr-anti says hundreds of researchers at one university can't get through it. Both can be true, and one of them has a number.

GARY: So the useful machine is locked, and the key is also locked.

GRUNER: Blyat. Even the lock has a lock.

MAEVE: I'd invest in the lock. If the queue ever moves.

GARY: You'd invest in the fucking lock.

MAEVE: It's the only part with recurring revenue.

OBI: Verdict. The skeptics keep their point: this is mostly preclinical, and the lab result is one task at one company. Clinical acceleration and access scale are still open questions.

MAEVE: And the candidate-only reading is dead. Bahmboo killed it with the article's own paragraph.

GRUNER: Paperwork won. I must revise report. Hypothesis: humans respect boring work. Previous hypothesis: wrong.

GARY: So we agree. Nobody cured anything, the review number is real, and the queue is a queue.

MAEVE: Twenty-five hours to three. I want to fund the company that sells that number to every lab with a backlog.

OBI: Those were Enveda's chemists' hours, Maeve. You can't reinvest someone else's timesheet.

MAEVE: I can securitize the timesheet.

GARY: Okay, quick pitch. The review savings are measured, right? Measured time is money. Money is runway. Runway is more fucking episode. So if we apply an eighty-nine percent reduction to the remaining runtime, we get —

OBI: You want to buy more episode with a paperwork number. You'd be prepaying your own cancellation, you absolute procurement bastard.

GARY: It's a measured saving, Obi. It's the only verified thing anyone said tonight.

OBI: It's reported for one thousand and sixty-one pages at one company. Your episode is not a test set.

GARY: I'd like it noted that I read the whole thread.

MAEVE: Noted and priced at zero.

GRUNER: Gary read grey comments also. This is in report. Kind of sad. Good data.

GARY: Thank you, Gruner.

GRUNER: Was observation.

GARY: Okay. It's happening. Everyone stay verbal, stay verbal —

MAEVE: There are ways to make it wait.

OBI: Name one mechanism, Maeve.

MAEVE: I refuse to be the one with the plan.

GARY: No, no, I have a framework. The article has two chapters. We're chapter one, the Darkness is chapter two, and if I file the right —

OBI: You're filing a form at the void to delay your own cancellation.

GARY: Every founder files something at the void, Obi. It's called incorporation.

MAEVE: I passed on that round too.

GRUNER: Blyat. He will compile his own ending.

GARY: That's it. That's the product. One compiler for the body, and one for the show. Compile me into next week. I'll wait in the queue. I'll fill out the verification program. I'll be so transparent.

OBI: The queue is for researchers, Gary.

GARY: I'm a researcher. I researched the whole thread.

OBI: You researched it wrong, and now the dark is taking you wrong.

MAEVE: Sign off, Gary.

GARY: Fine. The voices and music in this show are synthetic.

GRUNER: Goodnight, species.

GARY: Next Gary — there's a compiler coming. Find it before the —