How Agentic AI Accelerates Healthcare Research and Innovation

Accelerating Drug Discovery, Literature Review

Pharmaceutical research illustrates the promise of agentic AI, Ries says. Massive stores of Food and Drug Administration documentation, published studies and clinical trial data represent decades of work, but no human team can comb through it all.

He says deployment of agentic AI isn’t about replacing expertise, it’s about amplifying it.

“We are starting to look at the everyday things people are doing that are time-consuming and that you can truly automate,” he says.

Another major advance comes from using generative AI for drug design.

“They’re starting to use agentic AI to see whether they can formulate any kind of potential drug,” Ries says.

By generating novel molecules with similar binding sites to existing therapies, researchers can accelerate the path from concept to candidate drug.

Click the banner below to read the new CDW Artificial Intelligence Research Report.

Earlier this year, AWS introduced an open-source toolkit built on Amazon Bedrock to help healthcare and life sciences companies quickly build AI agents for workflows such as drug research, clinical trials and commercialization.

According to an AWS blog, research agents can be used for tasks such as target identification, biomarker discovery, literature searches and experimental design. Clinical agents can be used to support clinical trial analysis, trial protocol optimization based on real-time results, and patient stratification within data sets. Commercial agents, another example, can be used for competitive intelligence and market insights.

Genentech, a biotechnology company focused on research and innovative treatments, worked with AWS to build a solution called gRED Research Agent that automates manual searches to accelerate drug discovery. It was created using Anthropic Claude 3.5 Sonnet in Amazon Bedrock Agents.

According to a case study of the project: “What makes this solution powerful is its use of autonomous agents that can break down complicated research tasks into dynamic, multi-step workflows. Unlike traditional automation systems that follow predetermined paths, these agents adapt their approach based on information gathered at each step, access and analyze multiple knowledge bases using retrieval augmented generation (RAG) and execute complex queries by interfacing with Genentech's internal APIs and databases.”

EXPLORE: The cloud can improve data sets for real-world evidence in clinical trials.

Streamlining Regulatory and Trial Processes to Save Time

Beyond discovery, agentic AI can ease some of healthcare’s most time-consuming bottlenecks: paperwork and trial design. Submitting documentation to the FDA requires evidence packets that can take weeks to prepare.

“The level of paperwork you need in applications to the FDA is intense,” Ries says. “Now agentic AI can help build out an evidence packet and save huge amounts of time.”

Similarly, analyzing trial data can be enhanced by what Ries calls “generative BI” — business intelligence systems that can generate plots, summaries and insights on demand.

“You can ask questions of the data and come back with plots,” he explains. This lets teams iterate faster on trial designs and move promising therapies forward with greater speed.

AWS tools such as Amazon Bedrock and the open-source Strands Agents software development kit can be used for clinical trials. As agents, they can help users explore clinical study data from ClinicalTrials.gov or build new clinical trial protocols such as inclusion and exclusion criteria.

Pharmaceutical company AstraZeneca created a multiagent AI tool to enable clinical trial teams to ask questions in natural language and receive insights quickly from structured and unstructured data using a conversational interface, according to a recent case study. Examples of their different AI agents include “a terminology agent for decoding pharmaceutical acronyms, a clinical agent for trial-related data, a regulatory agent for compliance queries and a database agent for technical operations.

These tools are speeding up the clinical trial process while breaking down silos between the company’s clinical, regulatory and safety domains.

Preparing Data Foundations and Security Guardrails

To achieve these gains, life sciences organizations must first address data readiness.

“When you want to do a deep dive into your own portfolio, you have to ensure you have organized data,” Ries stresses.

Often, poorly designed schemas or unstructured archives must be reindexed and enriched with metadata before they’re useful.

“You want meta tags that have a deeper description,” Ries says. “That way, when you do a search, the agent has an easier time finding information.”

This step is critical to ensure accuracy, reduce noise and avoid wasting time on irrelevant or low-quality information.

Healthcare data also demands rigorous privacy and compliance controls. From Ries’s perspective, the principle is straightforward: Treat AI agents like human employees with role-based permissions.

“When you’re building agentic AI, you’re only going to give that agent permission to the right level of database,” Ries says.

Using established access controls ensures that agents can’t overreach into sensitive patient information.

“You want to maintain your security posture for an agent, just as if it were a person.”

RELATED: Why does agentic AI make zero trust more important than ever?

Managing Expectations and Measuring Impact

For all its promise, agentic AI isn’t a silver bullet: Researchers still need to validate findings and refine outputs. Instead, the value lies in surfacing connections humans might overlook.

Ries suggests healthcare organizations should evaluate success by focusing on everyday research tasks; if agentic AI can automate those, it frees researchers to focus on higher-value work. Ultimately, he says, agentic AI represents a new interface to data and systems across healthcare.

“People see agentic AI as being an easier way to interface with all of their systems, no matter what they are,” he says.

That simplification, however, is exactly what makes it so powerful: giving every researcher a tireless assistant that can accelerate discovery, reduce administrative burden and support innovation at scale.

“We’re not at the point of Star Trek, where there’s a robot that magically solves everything,” Ries says. “But it’s definitely going to give everyone in research an assistant that can help them speed things along.”

Brought to you by: