Drug discovery's bottlenecks sit everywhere, not just molecule design: Enveda's Claude review of 1,061 pages cuts human time by 89%
Synopsis
Anthropic's life sciences team reports that Claude has been used inside real drug programs at six organizations, on tasks reaching from molecule design to manufacturing-document review and regulatory filings. Discussion of AI in drug development has centered on proposing initial molecule designs, while manufacturing paperwork stayed manual, with chemists spending up to 75% of their time on records. In Enveda's test, Claude caught all 11 issues human reviewers had flagged across 1,061 pages of documentation and flagged seven more, cutting human review time by 89%, from over 25 hours to less than three; most of these results remain preclinical.
Interpretation
Work across drug development is starting to be handed to a single general-purpose model, from designing molecules to checking manufacturing documents, drafting regulatory filings and even planning clinical trials. The commonly discussed use of AI in pharma had been proposing initial molecule designs, while batch-record checks and filings stayed manual; at Enveda, manufacturing chemists spent up to 75% of their time on paperwork. With all of Enveda's manufacturing documents connected to Claude, a test over 1,061 pages found all 11 issues human reviewers had flagged and seven more, cutting human review from over 25 hours to less than three.
Inside each program the model is connected to that project's own data system and returns changes specific enough to synthesize, test or review. BigHat linked Claude to its laboratory data system Reccy, which proposed molecule changes; the lab built and tested about 1,010 variants in about 10 weeks. Among those variants, the best versions bound more than 10x more tightly, the cancer-targeting lead stayed in solution 18°C higher, and the longer-lasting drug lead clumped more than 100x more slowly.
Most of these results are still preclinical: the drugs have not been tested in people. Anthropic's own Mithril Medicine program took a candidate for a rare muscular dystrophy from idea to promising results in mice in under 18 months, and expects a drug ready to start human trials as soon as late 2028 if the candidates hold up. The team says it plans to extend the model's role into clinical trials as these programs advance and as it partners with organizations already at the clinical stage.
Perspective
The demonstrated scope is six partner drug programs plus an in-house program for rare childhood muscular dystrophy, covering molecule design, antibody-fragment optimization, manufacturing-document review, regulatory drafting and clinical-trial planning, with most results still preclinical. The team says it will extend the model's role into clinical trials as programs advance and will work with organizations already at the clinical stage; the in-house candidate is expected to be ready for human trials as soon as late 2028 if it holds up.
The research is described as still at an early stage and mostly preclinical, so the drug candidates have not been tested in humans. Partner impact is stated conditionally, on positive trajectories continuing, and the first skill built for The Column Group's new company Pleiades was evaluated on 45 target-disease pairs.