A Scientist + Claude, One Drug in 18 Months: Anthropic Just Acquired This One-Person Company

Anthropic has acquired a one-person biotech company. Founder Cody Tranbarger, known on Twitter as Professor Oak, used Claude to manage contracts, logistics, and experiments, coordinating 11 external labs, and within 18 months advanced an antibody therapy for LAMA2-related congenital muscular dystrophy from idea to mouse proof of concept.

Tranbarger joked on Twitter: after six years of anonymity, he finally exposed himself because his company was acquired. He said he spends most of his time tweeting and occasionally finds time to do drug R&D. Mithril Medicine was founded with support from Atlas Venture and was initially virtual, distributed, and ultra-lean. Later he found that the maturation of AI tools allowed this model to last longer, so Mithril itself became an experiment: how far can one full-time employee + a bunch of Claude instances + a group of part-time advisers go?

Anthropic calls this model Pipeline-in-a-Person: one scientist using Claude to advance multiple biological hypotheses simultaneously.

What Claude Did That Week

Anthropic's technical report includes several snippets.

At the start of handover week, the team internally completed the handover of two sets of cell experiments, and 11 questions were still pending with the CRO. Someone asked Claude to read the emails between the handover lead and the CRO and suggest next steps. Claude reached a conclusion in 96 seconds, and in 18 minutes produced a complete plan and 5 draft emails. It read 6 email threads, the handover folder, 2 slide decks, and 2 reports. Four business days later, 9 of the 11 questions had written answers, and the remaining 2 were left for pilot experiments.

On day 4, the CRO returned a 341-page data package. Someone forwarded it to Claude with two sentences, asking it to review, comment, and suggest next steps. Claude gave a preliminary judgment in 2 minutes 40 seconds, and 7 minutes 53 seconds later completed a visual synthesis of all 113 constructs, marking each construct as screen as-is, needs optimization, or on hold pending production standards. Every construct had a clear next step.

On day 7, someone challenged its recommendation. Claude defended it within 27 seconds. The human ultimately rejected one item and put 24 constructs on hold. Within 4 minutes, Claude updated the memo, work page, run order, and 3 unsent drafts; all affected dependent steps were adjusted in sync, and every change was marked for human confirmation. The whole process from challenge to new plan took 7 minutes 36 seconds.

That week, Claude drafted 17 emails; 12 were ultimately sent, and each was edited by a human. It additionally listed 41 details without written confirmation—such as vial fill volume, shipping conditions, recipients, and experimental setup—for a human to confirm or revise one by one.

The Division of Labor Has Changed

These details point to a more practical change: the way work is divided has changed. Claude takes on work that scales with volume: reading documents, calculating, drafting, and tracking how decisions propagate through all dependent steps. Scientific judgment is left to humans. The report includes a key description: a ruling on three arms determined the fate of 24 constructs. Screening scale is no longer limited by team size.

The experimental results were: severely ill mice that received the antibody treatment had about twice the grip strength and triceps strength of untreated mice, and more mice survived to the end of the study.

Of course, doubts exist. Some online commenters reminded: mouse experiments are not human clinical trials, and content generated by Claude needs human verification. Others said: CROs should pay attention to this.

Tranbarger believes that combining AI and biology will help drive cures for diseases, but mainly not through timeline compression. The low-hanging fruit he sees is orchestration and parallelization. Most drug R&D staff make only a few go/no-go decisions in their entire career, and spend a huge amount of the remaining time on email exchanges. Give them the ability to run multiple projects in parallel and let them spend their time on scientific judgment, and better decisions will naturally emerge.

Tranbarger is now continuing these two rare-disease projects at Anthropic while preparing to launch several new projects. Anthropic estimates that if the drug candidates perform well, they could enter human clinical trials by the end of 2028.

Mithril is not an isolated case. Over the past few months, Anthropic's life sciences team has also established drug discovery collaborations with six organizations: BigHat Biosciences, Bristol Myers Squibb, Divergence Labs, Enveda, Generate Biomedicines, and The Column Group.

One person + one model + part-time advisers has run a path that once required a team of dozens. This data point is worth noting.