Clinical trials are generating increasingly large amounts of data, but companies are under immense pressure to complete trials quickly and under budget. Could agentic artificial intelligence (AI) be the answer to getting the right patient to the right trial in a timely fashion?

Getting a medical treatment to market requires ever-increasing amounts of information and complex clinical trials. Advances in AI have made it feasible to incorporate more varied and high-volume information into clinical trials, from electronic health records and genomic data to imaging and continuous wearable-sensor measurements. These modalities may produce richer evidence, but can also make integrating, validating, and interpreting trial data more complex.

Likely due, at least in part, to this accumulation of data, trials are lengthening in duration. A report published by the IQVIA Institute last year states that the average time from Phase I to regulatory approval rose from 8.2 years in 2015 to 9.3 years in 2024, while trial start-up times also lengthened, from 4.8 months in 2020 to 6.1 months in 2024.

At the same time, companies are under pressure to get their products to market as soon as possible and to get more done with less money.

“Life sciences and healthcare continue to get more expensive, and no one thinks that is sustainable,” said Tom Doyle, CTO at Medidata, a New York-based forerunner in software and AI-based solutions for clinical trials.

“Every dollar we spend on a trial is a dollar not spent on a different trial. If we can run them more efficiently, then it gives us … more opportunities to bring therapies to market. We hear a lot of customers saying we want to run more trials, but on the same budget, and agentic AI is one solution for that seemingly intractable problem.”

AI agents are designed to pursue a goal by taking actions instead of simply producing an answer. They typically use generative AI, usually a large language model (LLM), to understand instructions, plan steps, and employ digital tools.

The use of such agents across many sectors has increased dramatically in the last couple of years. Before this, most computer systems required fairly rigid instructions to achieve any level of automation, but LLMs can now interpret broader instructions and achieve more complex tasks that require different steps. This has allowed the use of AI agents on a wider scale than before.

The use of automated or semi-automated AI in medical settings is more difficult than in other industries, as patient safety is paramount. However, advocates believe that these agents have the potential to revolutionize clinical trials. They can help with recruitment, complete standardized paperwork, monitor data collection, and coordinate simple administrative tasks, all of which can speed up trials and make them more affordable.

The right patient for the right trial

Patient recruitment is regarded as one of the main bottlenecks in clinical trials. If a study can’t enroll enough eligible participants, it may open late, run longer than planned, cost more, or even fail to answer its research question.

Grove AI was founded in 2024 by Stanford engineers Trân Lê and Sohit Gatiganti after seeing how difficult clinical trial recruitment could be. Lê’s own attempt to enroll in a trial for a chronic condition was slow and challenging, involving lots of complicated paperwork.

“When we saw that, and we saw the opportunity of these voice agents to accelerate labor and manual work, we definitely thought it was a no-brainer to bring these agents out in a safe and compliant manner,” said Lê.

The two engineers and their colleagues built Grove’s initial AI agent, “Grace,” to work across voice, text, and email, and automate patient engagement and recruitment. The aim was to decrease trial enrollment time from weeks to minutes, with the agent automating participant outreach, prescreening, scheduling, and follow-ups. This would lighten the load for trial staff and make the experience more accessible for patients.

“Typically, once you sign up and show interest in a clinical trial, before AI, it would take the trial sponsor seven to 10 days, if not more … to call you,” said Gatiganti. “Now, as soon as you click on the ad, our AI can call you, with consent, and say, ‘You expressed interest in this trial. Can I ask you a few questions to see if you’re eligible?’”

It did not take long for companies running clinical trials to realize the potential benefits of Grove’s offering. Early last year, the company said it had interacted with more than 70,000 patients and scheduled 7,000 in-person appointments. In January, Grove was acquired by Palo Alto-based Hippocratic AI, known for its patient-facing generative AI healthcare agents, where Lê and Gatiganti are now general managers of the company’s life sciences division.

“We’re probably one of the most experienced in the space,” said Gatiganti. “We’ve run recruitment for over 70 Phase II and III trials. We work with a variety of customers, from contract research organizations to large pharma, so we also bring that deployment expertise.”

Oncology-focused ConcertAI, based in Cambridge, Massachusetts, also uses AI to get the right patient to the right trial, although it has a somewhat different approach. It collects data from electronic health records and matches patients to trials, and helps trial sponsors find the right participants. Earlier this year, it launched an agentic AI platform to accelerate clinical trials.

“Where we’ve seen some of the biggest problems with studies … is designing a trial that doesn’t actually target real patients, and then because of that, what you thought would be easy … suddenly becomes incredibly difficult,” explained Robert Zambon, PhD, vice president of product management for clinical trial solutions at Concert.

“We’ve built up ontologies and knowledge graphs to let us have agents that understand oncology, understand how patients are treated, understand clinical trials, and have them work together to help us [ask], ‘Where are optimal patients, where are there gaps in care that can be addressed by a new therapy, and what should we highlight to an oncologist?’”

Concert’s ability to analyze records quickly is particularly suited to improving trial recruitment and helping cancer patients get treated in a timely fashion.

“We’ve seen uplift of 25–50% in the ability to recruit patients at a site, because you’re able to find ones [who] otherwise would slip through the cracks,” said Zambon.

Saving money and time

Medable, based in California, began as a digital and decentralized clinical trial platform, offering software for functions such as electronic consent, electronic clinical outcome assessments (eCOAs), telehealth, and remote data collection.

It launched its first generative AI offering in 2024 and, last September, moved into the agentic AI space with Agent Studio, which the company claimed was “the industry’s first agentic AI platform purpose-built for life sciences.”

“The way we’re thinking about these agents is that they are clinically focused teammates for people. … They can actually function and do things independently while you’re doing something else,” said Pamela Tenaerts, MD, CMO of Medable.

“It can help with about 90% of the jobs that a clinical research associate would do that are administrative in nature.”

Medable’s focus is more on trial monitoring and execution rather than on recruitment. This year, it partnered with the Tufts Center for the Study of Drug Development (CSDD) to assess how agentic AI could help make clinical trials more efficient.

The study results were released in August and showed that use of Medable’s trial monitoring agent could save around $4.4 million in running costs for a Phase II trial and around $5.6 million for a Phase III trial, with development time reduced by about 18 weeks.

They primarily used a calculation known as expected net present value (eNPV), a way of estimating what a drug development program is worth after accounting for variables. Medable and Tufts CSDD estimated that eNPV would increase by approximately $7.5 million for a Phase II trial and by as much as $21 million for a Phase III trial.

“This study really took a very deep and comprehensive look at study monitoring and all the areas where we could gain efficiencies, reduce costs, and drive speed,” explained Kenneth Getz, executive director and research professor at Tufts CSDD.

The results of this study are promising, but need to be backed up with longer-term data to confirm the proposed benefits.

“We do believe that there’s a quality improvement. It’s super hard to measure that right now because things are ongoing. You can’t look retroactively yet. But we do believe it’s dollars and hours saved, and then quality improvement,” said Tenaerts.

Medidata has been providing clinical trial software since it was founded in 1999 and, more recently, AI-based products. Like a number of companies in this space, it is also using agentic AI as a “virtual assistant” to help clinical trials run more smoothly and cost-effectively.

“A big area of focus, in particular around agentic AI, is the execution piece—how do we deal with this growing sophistication of trials, the growing ecosystem, the disconnected nature of a lot of the data and sources?” said Doyle.

“Another emerging area is agentic AI that is creating a more human-like experience. Simple things like notifications and reminders need to have more of a targeted approach. These are areas that we think are really important for creating a more patient-friendly experience in clinical trials, but also to reduce attrition.”

Hype or hope

The evidence in favor of agentic AI making clinical trials more efficient, patient-friendly, and cost-effective is accumulating, but it is still early days and large-scale roll-outs by companies are yet to occur.

“We surveyed about 200 leaders across pharma to get feedback on a number of different things, both [for] what we’re doing and industry trends,” said Doyle. “What that told us is [that] there was broad awareness; more than 90% of people were aware of agentic AI, but it still had relatively low adoption.”

A review article published in March concluded that most agentic AI studies in healthcare were exploratory, limited in scope, and lacked robust clinical validation. As the kind of AI agents that could help in trials have only been available for a couple of years, it will take time to collect enough evidence to back up trial improvement claims.

“At the moment, there’s still a lot of human oversight,” said Tenaerts. “Everything is audit traceable. Whatever actions the agent took, you can go back and see what that was and what the reasoning was. So it’s very open.”

Although AI’s accuracy continues to increase, all kinds of generative AI have the potential to “hallucinate” and mistakenly produce false claims. For example, a study of LLMs translating clinical trial eligibility criteria into database queries was published last year and showed that hallucination rates varied dramatically by model—from 21.1% to 49.5%, with an overall rate of 32.7%. While the study did not include AI agents in live trials, it examined a task that could underpin agents designed to find trial participants.

Hippocratic AI says it prioritizes patient safety when designing its agents. “All of our customers are responsible and know that they have to get patient consent as well as the script approved before we can go ahead,” said Lê. “We continue to be collaborative on leading the space in terms of policy and helping different entities understand how they can help us to better deliver safer agents.”

How agentic AI will shape the future of clinical trials remains to be seen, but it is certain to have a big impact. Lê said that it seems to be helping pharma companies who previously outsourced all trial management to better plan trials. “What we’re seeing now, especially with technologies like ours and others, is that pharmaceutical companies are really able to take more detailed control,” she said.

“The ability to do adaptive clinical trials is going to become more AI-driven than it is right now,” asserted Zambon. “I think over time, being able to find a patient is going to be tightly integrated into every system across the board. … Finding that needle in the haystack is going to become a lot easier as we move forward.”

 

Helen Albert is senior editor at Inside Precision Medicine and a freelance science journalist. Prior to going freelance, she was editor-in-chief at Labiotech. She was editor of The Biochemist magazine and blog, but also worked as a senior reporter at Springer Nature’s medwireNews for a number of years, as well as freelancing for various international publications. She has written for New Scientist, Chemistry World, Biodesigned, The BMJ, Forbes, Science Business, Cosmos magazine, and GEN. Helen has academic degrees in genetics and anthropology, and also spent some time early in her career working at the Sanger Institute in Cambridge before deciding to move into journalism.