Novo Nordisk And AWS Build Agentic AI Workflows For Drug Discovery

Novo Nordisk is pushing its AWS artificial intelligence programme beyond staff productivity and into the early science of drug development.

AI News reported that the deal names AWS as the Danish drugmaker's preferred cloud provider and strategic AI partner, with a London co-innovation hub as the operating centre.

The London site will place Novo Nordisk researchers in the same workstream as AWS technical teams, including cloud engineers, AI specialists and applied scientists.

Its role is to narrow the gap between computational screening and laboratory follow-up when potential medicines are still in early development.

The internal record already gives the project a large operating base.

More than 25,000 employees have used a Bedrock-based generative AI platform for chatbots spanning more than 2,500 use cases, and one deployment draws on about 140,000 documents while processing more than 26,000 prompts a month.

That internal record gives the new agreement a different starting point from a pilot announcement.

Novo Nordisk has already used AI systems for information retrieval, document drafting and non-regulated processes, so the London hub is aimed at research workflows where scientific validation and governance are harder than workplace automation.

Clinical-study paperwork offers one example of the claimed productivity effect.

AWS said a Bedrock system using Anthropic's Claude 3.5 cut the preparation time for some clinical documentation by more than 90%; the cited workflow moved from a 40-to-50-person effort lasting up to 15 weeks to a task completed in minutes by three people.

The generated material still goes through review by medical professionals.

The drug-discovery layer will use Amazon Bio Discovery, a service that exposes researchers to more than 40 biological AI models.

In Novo Nordisk's planned use, agents can help choose model sequences, analyse biological data, identify possible targets and rank candidate therapies before selected outputs move into synthesis and lab testing.

The research loop is the main operational change.

Instead of treating AI as a document assistant, the partnership is meant to move data from model selection to candidate ranking, physical testing, returned experimental results and further model refinement.

Bedrock applications are also expected to connect clinical, genomic and imaging data so findings from discovery work can shape trial design.

AgentCore supplies another part of the stack for deploying and running agents inside enterprise systems.

The service is designed to let agents use company data, link into existing software and execute longer task chains.

Novo Nordisk has not mapped those functions to named research projects, leaving the exact AgentCore deployment path undisclosed.

The partnership will also embed AWS Forward Deployed Engineers with Novo Nordisk teams as systems are built.

It extends a wider Amazon relationship that includes Amazon Pharmacy, Amazon Ads and One Medical, while the drugmaker has separately worked with OpenAI and Denmark's Gefion supercomputer on AI and data-intensive research.

The measurable test is whether the same AI stack that accelerated internal documentation can shorten research cycles while keeping clinical oversight intact.

Novo Nordisk chief executive Mike Doustdar framed the AWS work as part of the company's effort to speed drug discovery and expand AI use across chronic-disease programmes.