[Image licensed from Adobe Stock]
ConcertAI CEO Jeff Elton, Ph.D., has described oncology as an ideal field for such approaches. A single patient in oncology could potentially generate billions of data points across millions of records. The enormity of the data sprawl renders traditional approaches insufficient.
In an interview at NVIDIA’s GTC event, Elton described such AI agents as one might refer to employees—at least in the sense that they can collaborate with distinct roles. There is a kind of corporate-like hierarchy of AI agents, each with a distinct domain specialty with higher-level supervisor agents keeping tabs on specialists. For example, a “transaction agent” is specifically designed to “look at molecular diagnostic reports and actually extract the critical biomarker information,” formatting it into a usable condition. A higher-level “supervisor” agent “oversees that particular agent to make sure they do their work right.”
From data silos to integrated intelligence
Jeff Elton [Image courtesy of ConcertAI]
These specialized agents are already helping streamline clinical trial workflows, an area that traditionally saddled organizations with administrative burden and a central driver of billions of dollars of drug development each year. ConcertAI’s system can, for instance, analyze data from tens of thousands of cancer patients, rapidly spot treatment patterns, and suggest optimal clinical trial designs.
The foundation for the ConcertAI and NVIDIA agentic partnership was revealed in June 2024, when ConcertAI announced that it was joining forces with NVIDIA to provide the computational muscle and AI expertise to realize their agent-based vision.
Economies of agentic scale
Elton emphasized that transparent reasoning capabilities are crucial for these AI systems. “When you add that reasoning component, you can actually say, show the work and show the logic of why this is the answer.” Such transparency is indispensable in healthcare contexts: “If you’re in a field, like medicine, and you’re asking for recommendation logic… that’s pretty high value. It shows me the logic about how I arrived at that that’s pretty compelling to the team.”
The efficiencies gained through these agent-driven workflows address the challenges pharma companies face. “Unfortunately there’s a lot of administrative activity, supporting analysis—classic management analysis, sometimes looking at what proportion of their total time is the value-added key decision time in terms of what you’re doing,” Elton said. ”And I think one part that agents can contribute to a lot is taking a lot of the less value-adding time that still requires humans today, and actually doing automation and higher performance.”
Toward a job market for agents
Rather than simply automating isolated tasks, agents could potentially do more than augment human workflows. There could be a sort of job market for agents as well. Their specialized nature means they can be deployed where they deliver measurable value. In pharma, they could address genuine clinical and research challenges rather than support generic automation. This points to a future where organizations might ‘hire’ specialized AI agents for specific functions, similar to how they currently engage consultants or specialists. “Our view is agents, where people will be marketing them, will be a valuable part of a lot of solutions, but people will be paying for them when they have that specialization and that differentiation,” Elton said.
Elton noted how economic pressures are changing pharmaceutical operating models: “Pharma is still trying to understand how they can lower the cost of bringing new medicines forward,“ he said. These agents’ potential to optimize individual trials could evolve into something even more valuable—the ability to analyze patterns across multiple studies to formulate better clinical trial designs industry-wide. When asked about agents that could review and optimize across dozens of similar trials (or more), Elton was upbeat: “I actually think that could work quite well. One of the things we’re going to start doing is we’ll start bringing in more guidelines, more features,“ he said. “So we’re bringing in what we would consider to be highly legitimate knowledge baselines.” Such baselines could help raise the floor of clinical trial design quality across the industry, ensuring that even baseline studies incorporate validated best practices and evidence-based protocols.
Filed Under: machine learning and AI
