Novo Nordisk Puts Claude Science Into Drug Discovery

Novo and Anthropic said on Wednesday that they will work together on drug discovery. The Danish company will test Anthropic’s Claude Science workbench inside selected research and development workflows, and will use Anthropic’s frontier models to support AI-driven software engineering across the business. The announcement, issued from Bagsværd, discloses no financial terms, no target molecules and no milestones.

Claude Science is the research workbench Anthropic launched in beta on 30 June: sixty-odd preconfigured scientific skills, connectors into structural and sequence databases such as PDB and UniProt, an NVIDIA BioNeMo integration, and a reviewer agent that checks the citations its own outputs produce. It was pitched at exactly the kind of work Novo is describing — literature synthesis, biological reasoning, and the long tail of computational chores that sit between a hypothesis and an assay.

What the two companies have actually committed to is narrower than the framing suggests. The stated plan is to jointly identify drug-discovery problems that Novo’s scientists and computational teams already run into, then build targeted solutions for those specific workflows. The release describes an initial aim of testing Claude Science where the companies expect their combined capabilities to have the greatest impact, with data governance and human oversight written in. That is a pilot with an ambition attached, not a research programme with a deadline.

The quieter half of the deal may be the more consequential one in the near term. Novo is also putting Anthropic’s models behind software engineering company-wide — agentic coding, in other words, across a manufacturer that employs more than 67,000 people. Pharmaceutical companies run enormous amounts of internal software around trials, manufacturing and regulatory submission, and that is a domain where current models demonstrably work today, unlike biological reasoning, where they are still being evaluated.

Novo is not betting on one lab. In May it signed a strategic partnership with OpenAI covering discovery, clinical trials, manufacturing and corporate operations, with integration targeted by the end of this year. Chief executive Mike Doustdar described the Anthropic collaboration as "another testament to our ambition to become the world’s most AI-driven healthcare company, building on Novo’s current AI initiatives with other technology partners" — an unusually direct acknowledgment that the company is running frontier vendors in parallel rather than picking one.

The timing is not incidental. Novo is in the middle of the hardest stretch in its recent history: roughly 9,000 job cuts announced this year, about 11% of the workforce, aimed at saving 8 billion Danish kroner a year, with around 5,000 of those losses at home in Denmark, according to Pharmaceutical Technology. Eli Lilly has taken the lead in obesity drugs, the company has pre-announced steep price cuts on Wegovy and Ozempic for 2027, and it shortened its name to Novo two days before this announcement. An AI productivity story arriving in the middle of a cost programme is a familiar shape.

Dario Amodei used the release to repeat his standing claim that AI can "compress a century’s worth of biological and medical breakthroughs into a decade." Nothing in the announced scope tests that. Testing a workbench on selected workflows at one drugmaker is a reasonable thing to do and a very long way from compressing a century, and the distance between those two sentences is the part worth watching.

The check on all of this is unusually slow and unusually clear. Either something Novo files in the coming years traces back to work that could not have been done before these tools existed, or the collaboration ends up as a very capable set of internal engineering assistants. Both are useful outcomes. Only one of them is the one being advertised, and neither company has promised to tell us which it got.