By Deborah Borfitz 

October 8, 2025 | Clinical research associates (CRAs) may soon be working alongside an artificial intelligence (AI) agent offering relief from some of the drudgery of the job, including crunching data from a dozen or more data sources in search of meaningful insights and actionable information. Medable, the leading clinical trial platform provider, has just launched Agent Studio, the first agentic AI platform purpose-built for life sciences that comes with an out-of-the-box CRA agent. 

Agent Studio has been described as a “Lovable for life sciences,” says Andrew Mackinnon, senior vice president and executive general manager at Medable. But instead of building websites using text prompts, Agentic AI is being leveraged to remove unproductive “white space” in clinical trials and make them more site- and patient-friendly. 

Agentic AI is already being used for research purposes, notably to interpret data in the context of lab bench discovery, he says. Other “agent studio” platforms also exist to automate tasks and streamline business processes and, like Medable’s Agent Studio, are powered by large language models. 

The difference here is that the platform is provided by an “experienced vertical company” specializing in clinical trial execution, says Mackinnon. That’s an important distinction given the complexities and challenges of working in the extremely regulation-heavy life sciences. 

Regardless of one’s role in a clinical trial—e.g., CRA, data manager, medical monitor, or safety writer—an “agentic teammate” can be created to pull data together and analyze it to suggest next best actions for the creative human-in-the-loop to take, Mackinnon explains. With Agent Studio, AI agents can be built-to-suit in a matter of days if not hours and ready for real-world deployment. 

The CRA agent is but the first in a series of off-the-shelf software that will ultimately be running on Agent Studio, he adds. It was the chosen starting point because of the significant challenges CRAs face due to the variety of systems and protocol structures used in clinical trials.  

Beyond the CRA agent, other AI systems likely to have high proof of value include those specific to protocol design optimization and data management monitoring, says Mackinnon. Both have been shortlisted for development by Medable. 

Agent Studio could have a “huge impact across almost every area of the clinical research life cycle,” he says. The platform allows companies to build agents tailored to their own specific use cases and the potential avenues for exploring the agentic approach are “essentially limitless.” 

Two-Step Process 

As Mackinnon has learned firsthand from numerous customers, it can be quite difficult for customers to build an AI agent for clinical research on their own that can be validated and adopted at scale. Internal teams might use ChatGPT to build a potentially helpful tool, but it invariably gets shelved by their systems validation and QA compliance colleagues for lack of evidence that it works as intended.   

Agent Studio has built-in validation processes for different use cases that can be provided in a GxP (Good Practice) compliant validation package—crucial evidence used to support submissions to the U.S. Food and Drug Administration and demonstrate that a system meets regulatory requirements for product quality and safety. The European Medicines Agency as well as Good Clinical Practice (GCP) guidance put out by the International Council for Harmonization also call for that kind of validated computer system approach, Mackinnon notes. 

The out of-the-box CRA agent is nothing like the standard eClinical systems being used in clinical trials, he continues. It more closely resembles “a brand-new, fresh CRA .... [with] a foundational grounding in scientific knowledge” and how to maintain ethical and regulatory compliance. 

The requirements of research, and ethical and regulatory guardrails, are built into the CRA agent. “When we then implement it into a particular customer’s ecosystem, we add specialist knowledge on top of that,” says Mackinnon. To that end, the agent can be given access to the company’s clinical systems for, among others, clinical trial management, electronic trial master file, electronic data capture (EDC), interactive response technology (IRT), and randomization and trial supply management. 

The agent thereby receives additional training on a company’s standard operating procedures (SOPs), monitoring plans, and protocols so it can function in accordance with those SOPs, work instructions, and practices, he says. The same two-step process will be applicable to many of the other out-of-the-box agents Medable has on the drawing board. 

Change Management 

Customization of an out-of-the-box AI agent is not a particularly complicated process, says Mackinnon. “The contextual knowledge that can be given is really just a case of providing access to the documents.” The knowledge systems structure of Agent Studio gives each agent access to different folders of information. 

When it comes to integrating with various clinical systems, Medable works under the Model Context Protocol describing the access process for agents into different systems, he explains. “They’re able to read the information and to take the actions that they have been given permission to take.” 

This step can take a little time to build, Mackinnon adds, although Agent Studio has numerous built-in connectors to popular, pan-industry systems—Veeva, Oracle, and several IRT systems among them as well as enterprise platforms such as Microsoft 365 and Salesforce. But the “time to value is going to be very quick.” 

The key to enterprise-wide adoption of AI agents is much less about overcoming the technical challenges than the change management process, he stresses. It is also where Mackinnon and his team are putting their focus, homing in on the ability of agentic AI to eliminate a lot of the administrative burdens on CRAs so they can do their job more effectively. 

This, in turn, supports sites in the operationalization of a study, points out Mackinnon. “At the end of all the number crunching, pulling data from different systems, trying to align it, [determining] what’s working well and what’s not working well ... you’ve got a human, the CRA, interacting with the human study team at the site solving problems they are having delivering on that particular protocol.”  

Thanks to the CRA agent, the human CRA knows what conversations are needed with the study investigator, study nurse, and research coordinator and has time for those discussions, he says. The result should be protocols completed more efficiently and successfully. 

It could also help Medable move closer to its vision of increasing the number of drugs coming to market, says Mackinnon. “One of the ways we do that is enabling companies to run more trials [by] being able to do more with the resources that they have.”  

User Interface 

The way users interact with the CRA agent is flexible and not solely a chat interface, says Mackinnon. It presents a natural language summary of happenings in the last seven days and highlights “things you need to pay attention to” based on company- or study-specific criteria the agent was instructed to prioritize, such as safety reporting. 

CRAs use clickable tiles to learn what the agent has found, including the areas where the data is pointing to an issue and flagging sites that are outliers compared to peers within the clinical trial—in terms of adverse events reporting, for example. If there is a mismatch between concomitant medications and adverse events, the CRA will probably want to raise a query and have the agent make that request for clarification, he says. 

Based on its monitoring of site enrollment, the agent might also suggest when drug inventory levels need to be increased, continues Mackinnon. The agent will also know if there is a shipment delay, or if a site has failed to confirm receipt in the IRT, so the agent can direct the CRA to take steps to quickly resolve the issue. 

Importantly, the CRA agent also has a view across different data sources to monitor protocol compliance, he says, so that the CRA can proactively intervene rather than perhaps learning of the deviation six to eight weeks later during an on-site visit. Failure of sites or patients to comply with data-recording requirements, leading to under-powered studies, is an all-too-common reason for study failures. 

The CRA agent would be trained anew at the start of every clinical trial using standard study documents such as the protocol and monitoring plan, says Mackinnon. In addition to understanding what to look for, the agent is also monitoring “all the other things that don’t make sense within the study, which provides CRAs with additional information they wouldn’t necessarily have had ... [if] they were just manually reviewing each individual system.” 

It perhaps comes as no surprise that Medable did a lot of discovery work with CRAs on the front end to understand their challenges, problems, and activities where they would welcome some support. They also were the first audience for a demo of the CRA agent, says Mackinnon, and they were excited about the prospect of leaving behind the “really difficult, dull, boring, manual data manipulation tasks they used to have to do.”   

Study sponsors and contract research organizations (CROs) have “huge plans for agents and no real way to start ... prototyping and building some of those ideas and very quickly take them into implementation, prove the value and then start to scale them across their organization,” Mackinnon reports. They commonly describe Agent Studio as “game-changing” by giving them a way to do that—all from the same platform. 

Marking Progress 

Medable can provide sponsors and CROs with any engineering support needed to implement individual AI agents, but companies can also access the platform and independently build and deploy those systems on their own, Mackinnon says. The broader Agent Studio platform is compliant with GCP regulations as well as the European Union’s General Data Protection Regulation and the U.S.’s Health Insurance Portability and Accountability Act, all involving role-based user permissions. 

Agent Studio can “sit next to or on top of” the broader Medable platform, or all on its own, Mackinnon continues. Medable-collected electronic consent forms and clinical outcome assessments are important data sources used with the CRA agent and will also be important in future development of a data management agent. But the use of other clinical systems for these data-collection tasks, and multiple others beyond Medable’s business interests (e.g., EDC, lab and imaging results) doesn’t preclude the use of Agent Studio.

It’s early days, he notes, but Medable is already actively working with customers looking to utilize AI agents in their clinical studies. Many companies have mandates from the C-suite to adopt the agentic approach. Demonstrating proof of value is a critical part of the process, says Mackinnon. Early use cases will be capturing key performance indicators to measure pre-defined success criteria. 

Meanwhile, Medable will be tapping the “treasure troves” of existing information companies have been sitting on to create a new generation of out-of-the-box agents, he adds. This retrospective data includes insights about protocol successes and failures, operational wins, and regulatory feedback. 

“Industry has been crying out for innovation for a long time,” says Mackinnon, but progress to date has fallen short of genuinely serving patients and sites better. He therefore encourages people in the industry to challenge their cherished but dated ideas about how clinical research must be done to make room for some tireless new teammates.