A 14-minute video from David Shapiro. He opens with his account of Anthropic buying a one-person company, Mithril Medicine, then walks through an Anthropic report on speeding up drug development with Claude. In the second half he uses that report to argue for roughly a million times more compute, which he links to personalized medicine and longevity escape velocity. The acquisition details are the creator's account of events, as he tells them in the video.
Shapiro says that on October 27th, 2026, Anthropic announced it had bought Mithril Medicine, founded by Cody Tranbarger, who used Claude to run experiments with 11 outside labs. He then summarizes five examples from an Anthropic report called Early Progress on Accelerating Drug Development with Claude. His argument is that the bottleneck in medicine is the number of expert teams, and that compute can replace much of that limit. He says a million times more compute is reachable in about ten years at the historical growth rate.
What it is. Shapiro says Anthropic bought a company made up of exactly one person. Mithril Medicine had been developing treatments for two rare forms of muscular dystrophy. According to Shapiro, founder Cody Tranbarger used Claude to organize the contracts, the logistics, and the experiments with 11 outside labs, and handed many other parts of running the company to Claude.
Why it matters. In his telling, Mithril took a possible treatment from an idea to a promising result in mice in under 18 months (00:24 to 00:34). He calls this what one person can do with AI today. He says Tranbarger now works at Anthropic and plans several new programs, and that Anthropic expects one drug candidate to be ready for human trials as soon as late 2028.
How to apply. Treat this as the creator's summary of a report, and read the primary source (00:57) before repeating any claim. Then use it as a prompt for thinking about which work in your own field could be delegated to an AI-assisted solo operator.
What it is. Anthropic's Life Sciences team worked with six organizations, and Shapiro says the point is that the bottlenecks are spread across the whole process. He covers five. Generate Biomedicines used Claude Code to propose a design about 70% shorter for part of a CAR T-cell therapy, and in lab tests the cells made about 10 times as many receptors. Divergence Labs had Claude generate 35,000 antibody designs in about a month, and three of 20 designs that pulled IgE off cells appeared to beat standard medications. Big Hat Biosciences tested more than 1,000 versions of two candidates in about 10 weeks.
Why it matters. The fourth and fifth examples are about paperwork and planning. At Invata, chemists spent up to 75% of their time on records, and Claude caught all 11 problems human reviewers found plus seven more across 1,061 pages (04:25 to 04:37). A regulatory document took four hours instead of about three months. Bristol Myers Squibb uses Claude to plan trials and estimate their chances of success.
How to apply. Look for your own slowest step, whether design, testing, documentation, or planning, and test AI on that one step first.
What it is. Shapiro asks whether this is mostly lab tests and mice right now, and answers yes, adding that Anthropic says so itself. He says most of the research has not reached human trials and that plenty of drugs that look great in the lab never reach patients.
Why it matters. His claim is that the point is not any one drug. The point is how many more ideas a small team can try, and how fast (05:25 to 05:31). That reframes progress as a change in the rate of experimentation, not a single breakthrough.
How to apply. When you read AI success stories, separate early-stage results from proven outcomes, and ask what is actually being measured. At 05:46 he also pauses for a message saying he takes no corporate sponsorships, his channel has been demonetized by YouTube, and he is supported by his community through Patreon, Substack, or Twitter. That is context about the creator, not part of the argument.
What it is. Shapiro says people ask what we would do with a million times more compute, and his answer starts with personalized medicine. He tells the story of KJ Muldoon, born in the summer of 2024 near Philadelphia with CPS1 deficiency, a rare genetic disorder. A team at Children's Hospital of Philadelphia and Penn Medicine designed and manufactured a gene editing treatment for KJ alone within six months, with doses beginning in late February 2025.
Why it matters. KJ went home after more than 300 days in the hospital. Shapiro notes that the FDA proposed a new way to approve such treatments in February 2026 (07:52). The catch is staffing: about 30 million Americans live with a rare disease, more than 10,000 rare diseases are known, and only about 500 have a treatment (08:16 to 08:27).
How to apply. Notice where a process is limited by scarce experts, not by the underlying science. His claim is that this is where compute can substitute for headcount (08:41 to 09:06).
What it is. Shapiro argues every ailment, including cancer, heart disease, and aging, deserves treatment personalized to the individual, because every tumor and every body handles drugs differently. To make a million-fold increase concrete, he takes Divergence's 35,000 designs a month and multiplies by a million, getting 35 billion designs a month, or more than four for each of the roughly 8 billion people on earth.
Why it matters. He says a million is not science fiction. By his account, the compute used to train the biggest AI models has grown four to five times every year since 2010, 16 years running. At four times a year, ten years gives about a million-fold increase (10:17 to 10:41).
How to apply. Check the arithmetic yourself: the claim rests on a growth rate continuing for a decade. Ask what would have to hold true, such as supply, cost, and demand, for that rate to continue, and treat his forecast as an argument, not a fact.
What it is. Shapiro credits the gerontologist Aubrey de Grey with the idea. Longevity escape velocity is the point where medicine gives back more than one year of remaining life expectancy for every calendar year that passes, so you are no longer racing the clock.
Why it matters. He says the world record for women's life expectancy has risen by about three months per calendar year for 160 years, while escape velocity needs 12 months. That is four times faster than now, and the gains must reach people of every age alive today (11:15 to 11:42). He describes aging as many kinds of damage piling up at different rates, so each person must be measured and repaired repeatedly for life.
How to apply. Use the framing to evaluate health technology claims: ask whether a treatment adds time faster than time passes. Shapiro concludes that every step, reading biology, designing treatment, and running experiments, runs on compute (12:18).
Shapiro closes by saying that when he calls for a million times more compute, he means every person on earth gets a team like KJ's. He says the science is already working and regulators are making room, and that what is missing is compute. The final minute and a half of the file (from about 13:05) contains unedited retakes, where he stops and restarts lines, so there is no further content there.