The most useful number in Anthropic’s new drug development report isn’t about molecules. It’s that chemists at Enveda spent up to 75% of their time on paperwork, and Claude cut one review from over 25 hours to under three. That is less glamorous than AI-designed drugs. It may also matter sooner.
Anthropic announced results from several months of partnerships with six organizations: BigHat Biosciences, Bristol Myers Squibb, Divergence Labs, Enveda, Generate Biomedicines and The Column Group. Anthropic has also started its own programs, including ones for rare childhood forms of muscular dystrophy. The pitch is that bottlenecks are spread across the whole pipeline, so AI should be involved in all of it, not just molecule design.
What the case studies found
Four of the six have published numbers. They are early and come from the companies themselves, so read them as promising, not proven.
- Generate Biomedicines: Six scientists used Claude Code to design a shorter piece of a CAR T-cell therapy, aimed at eventually treating solid tumors. One design was about 70% shorter and let immune cells make roughly 10 times as many receptors in lab tests.
- Divergence Labs: Claude, largely autonomously, generated about 35,000 antibody designs in a month for a food allergy drug. Early tests found three designs that bind IgE and 20 that dislodge it.
- BigHat Biosciences: Claude, connected to BigHat’s lab data system, proposed changes for about 1,010 variants tested over 10 weeks. The best bound more than 10x more tightly, and one clumped over 100x more slowly.
- Enveda: Claude caught all 11 issues human reviewers found, flagged seven more, and drafted a regulatory document in four hours that would take about three months by hand.
Why the BigHat result stands out
Partway through, the team suspected the standard stability and clumping metrics were misleading. Claude and an Anthropic researcher reanalyzed the raw data and proposed better ones, and the main predictions held. That’s closer to scientific judgment than to generating candidates, and it’s what separates this from a pure screening tool.
How it compares
AI drug discovery isn’t new. Isomorphic Labs, Recursion and Generate itself have built models for protein and molecule design. Anthropic is taking a different route: a general-purpose model applied across design, operations and documentation. That’s a bet that breadth beats a specialized model at any single step.
Does it matter?
Partly. The hit rates and lab improvements are encouraging, but none of this is a drug in patients. Hits are not cures, and the long, costly stages are still ahead. The Enveda numbers are the most concrete, since time savings on documentation are easy to measure and don’t depend on biology cooperating.
Watch The Column Group’s new company, Pleiades, which plans to run far more drug programs than a small team normally could. If that works, it changes how biotech investors price risk. If it doesn’t, we’ll learn where Claude’s limits actually are, which Anthropic says it wants to find out anyway.
