AI Agents Move Drug Discovery From Search Toward Science

AI agents are moving drug discovery beyond isolated experiments.

Novo Nordisk is expanding its use of AWS artificial intelligence tools across drug discovery. The work includes target identification, therapy design, and research workflows.

Under an agreement announced recently, AWS becomes Novo Nordisk’s preferred cloud provider and strategic AI partner. The arrangement gives a familiar cloud relationship a more ambitious job.

The companies created a co-innovation hub at Novo Nordisk’s existing London facility. AWS engineers and Novo Nordisk scientists will work together there.

The hub aims to shorten the path from identifying a drug target to a first human dose. That is the part of drug development where time becomes expensive.

Thilde Hummel Bøgebjerg, executive vice president of Enterprise IT & Quality at Novo Nordisk, described the goal clearly.

“AI has the potential to transform how we discover, develop and deliver medicines, but real impact comes from combining advanced technology with deep scientific expertise and a clear focus on patients.”

Novo Nordisk Builds an AI Operating Layer

Novo Nordisk already has evidence of broad internal use. More than 25,000 employees have used a generative AI platform built on Amazon Bedrock.

Those employees created chatbots covering more than 2,500 use cases. One deployment uses about 140,000 documents and processes over 26,000 prompts each month.

These figures show an enterprise platform moving beyond a small pilot. They also show the dull machinery behind AI adoption—documents, prompts, and workflows.

Novo Nordisk is not limiting AI to research teams. The agreement covers AI agents for target identification, therapy design, and research workflows.

H1 is adding another layer to the company’s clinical development work. H1 announced a partnership with Novo Nordisk to develop AI-ready data and AI-powered clinical development workflows.

H1 will acquire rights to further develop StudyHub. Novo Nordisk built the digital and AI-enabled platform for clinical development.

H1 is an AI-powered platform for identifying and engaging the right doctors. Its work spans pharma, health plans, health systems, and technology companies.

Its Doctor Graph represents physician identity, expertise, relationships, and behavioral signals. That structure gives AI workflows a more useful map of medical networks.

H1 combines healthcare data with agentic AI workflows. The system supports drug development, medical and commercial engagement, care navigation, and provider data intelligence.

It also supports network intelligence. In this age of AI, clinical trials need a new operating model.

“In this age of AI, the way clinical trials should be executed needs to change,” said Ariel Katz.

Mishal Patel called H1 the right partner for the work. Patel cited its domain expertise, AI-ready proprietary data, and AI technology.

The Larger Bet Is Scientific Reasoning

The push reaches beyond one pharmaceutical company. Amazon is redesigning part of a massive AI data center campus in rural Indiana.

The redesigned site will become a sprawling cluster of powerful computers. The project sits inside a broader initiative called “AGI Pivot.”

That initiative supports Amazon’s AGI organization and its future in-house AI models. Data centers remain the quiet foundation beneath grand theories about machine intelligence.

The scientific case for AI agents starts with a stubborn problem. Many fields cannot create perfect datasets for every question.

In 2024, Demis Hassabis and John Jumper of Google DeepMind received part of the Nobel in chemistry for AlphaFold. AlphaFold predicts three-dimensional protein structures.

It learned from thousands of experimentally measured shapes. The Protein Data Bank contains roughly 170,000 experimentally validated protein structures.

Creating that database took 53 years of international cooperation. It also required roughly $21 billion worth of experimental work.

Most experimental sciences face similar data limits. Variability and measurement challenges make comparable datasets difficult to generate.

Scientists have traditionally handled uncertainty by combining methods and applying judgment. Perfect datasets were never part of the bargain.

AI agents offer a different approach. They are reasoning engines that use digital or physical tools during discovery.

They can mimic the iterative process of scientific research. They can also log every move and create exact records for replication.

Agents can record a lab’s entire scientific history. A centralized repository could preserve institutional knowledge instead of letting it vanish with staff turnover.

Google’s AI Co-Scientist offered a glimpse of that model. Announced in May, it determined that resistance genes hitch rides on bacterial viruses.

That matched a conclusion reached after a decade of wet-lab work. “The hypothesis was correct,” the AI Co-Scientist said.

An agent that reads a thousand papers in an hour could change research timelines. One that designs 500 molecules and learns from failed tests by morning changes them further.

AlphaFold answers specific questions. Agents are generalists that model the human process of discovery.

Eric Schmidt, former CEO of Google, co-founded Schmidt Sciences in 2024 with his wife Wendy. Suhas Mahesh leads its AI for Science work.

Maya Levin is a researcher at Schmidt Sciences. The emerging contest is no longer just about better models.

It is about connecting models to data, tools, experiments, and institutional memory. Drug discovery has gained a new machine room.

Stay connected via Google News

Follow us for the latest AI News