Prasanna Rao, Chief Products and Innovation Officer at Saama, explains how Agentic AI — combining workflows and large language models — is set to streamline clinical development, reduce cycle times, and empower human decision-making in pharma. He outlines its applications in data management, trial design, and regulatory compliance, offering a glimpse of a future where humans and AI collaborate seamlessly to accelerate drug development.

From RPA to agentic AI

Looking back a decade, Prasanna Rao recalls his time at Pfizer, when robotic process automation, or RPA, was revolutionizing repetitive tasks. One early use case involved reconciling adverse event data between the safety system Argus and an electronic data capture system. In clinical trials, when a patient experienced a severe adverse event — such as a heart attack — the law required that it be reported within 24 hours in the safety system. At the same time, the electronic data capture system tracked similar events. Reconciling these two systems quickly and accurately was a manual and time-consuming task.

“RPA was doing something very unique: log into two systems as a bot, bring in data, and then let a human decide the next best action,” Rao explains. “It was saving time for knowledge workers and reducing the risk of errors in critical safety data.”

Fast forward ten years, and that same task can now be handled with agentic AI, which combines a workflow with a large language model capable of reasoning. The agentic AI system can identify that a “heart attack” and a “myocardial infarction” reported in different systems are logically the same event. It then flags the duplication or reconciliation needed and recommends the next step to a human data manager. The decision-making authority always remains with the human, but the agentic AI performs reasoning and analysis at a scale and speed that RPA could never achieve.

“An agentic AI is nothing but a workflow and a large language model that can reason and suggest the next best action for a human. There’s nothing the agentic AI does independently — the human is always in the loop,” Rao says.

Why now?

The renewed interest in agentic AI is driven largely by technological advances and its growing ubiquity. Tools like ChatGPT have made large language models familiar to the wider workforce, with functionalities that extend far beyond search or planning.

“Technology is far advanced. ChatGPT has become ubiquitous. You go in and there’s an agent button — suddenly the agent can connect to anything and do more than you’d expect,” Rao explains. “Last year, ChatGPT could plan an itinerary. Now, the agent can connect to Expedia and start making bookings. That leap from planning to acting is what agentic AI brings to pharma.”

In the context of clinical trials, this opens the possibility of persona-based agents tailored to specific roles, such as data managers, clinicians, or medical monitors. Multi-agent orchestration can connect workflows across disparate systems, reducing redundancy, saving costs, and improving coordination.

“Every single role in clinical trials could have an agentic AI. Orchestrating a multi-agent process can save costs and connect the dots across systems.”

Applications in data management

Saama has been exploring AI in clinical development for many years. Rao cites a striking example from the COVID-19 pandemic: the company helped Pfizer accelerate vaccine development, saving a full month in the process.

“Every hour mattered, every day mattered — lives were at stake,” he says. “Since then, we’ve been tracking data management closely, moving from pure machine learning to natural language processing, and now to agentic AI.”

In practice, agentic AI can function as a Junior Data Manager, performing tasks like reconciling data entries, detecting inconsistencies, and making intelligent recommendations to senior data managers. This frees up human experts to focus on decisions that require experience and judgment.

“What if tomorrow you could have an agentic AI acting as a junior data manager? Agents can run tasks, reconcile data, write emails, follow up with sites — all while the senior data manager focuses on decision-making,” Rao explains.

The potential efficiency gains are significant. Cycle times for database lock and reporting can be reduced, while human oversight ensures accuracy. Rao estimates a conservative 25% reduction in cycle time is achievable through agentic AI but emphasizes that the real value lies in the combination of time savings, cost efficiency, and enhanced accuracy.

“Agentic AI never sleeps — it can run for 24 hours, bringing both cost and time efficiencies. That’s something a human data manager can’t do,” he says.

Impact on trial efficiency, data quality, and decision-making speed

Rao notes that the combination of speed, scale, and reasoning offered by agentic AI could have far-reaching consequences for trial efficiency, data quality, and decision-making.

“Cycle time will reduce — conservatively, I see a 25% reduction. And because agentic AI can run 24/7, you also get cost efficiencies,” he says. “The implications for patient outcomes are substantial: faster, more accurate trials mean faster access to therapies.”

He points out that this is not just about automation; agentic AI can also enhance decision-making quality. By analyzing complex datasets, linking protocol elements to endpoints, and providing actionable insights, agents can help clinical teams make better-informed choices earlier in the development process.

Building trust and regulatory compliance

One of the biggest challenges in adopting AI in pharma is trust. Rao stresses that transparency, auditability, and regulatory compliance are essential for agentic AI to gain acceptance.

“Adoption is equal to trust in life sciences. You have to see what the agentic AI is doing — the chain of thought, the plan cycle — and intervene, when necessary,” he says.

Agentic AI systems can be trained to follow standard operating procedures rigorously, ensuring they operate within defined parameters. Rao highlights the example of enterprise ChatGPT adoption in pharma as a blueprint: initially banned over fears of data leakage, it was later introduced in secure, controlled environments, allowing safe use of proprietary data.

“The same approach works for agentic AI,” Rao says. “You define the guardrails, ensure human oversight, and you can safely deploy it while meeting regulatory requirements.”

He adds that the EU AI Act, emerging guidelines on AI passports, and robust internal SOPs are helping pharma companies build frameworks for explainable, accountable AI. The ultimate goal is patient safety: any AI or agentic system must operate in ways that do not compromise clinical trial integrity or participant wellbeing.

Human roles and barriers to adoption

Despite the promise of agentic AI, humans remain central. Rao emphasizes the importance of careful use-case selection, starting with high-feasibility, high-impact scenarios.

“We analyzed 150 use cases and focused on those that were technically feasible and offered maximum impact. You string them together systematically, and suddenly agents solve meaningful problems, building trust along the way,” he explains.

The digital protocol initiative at Saama illustrates this principle. By codifying complex clinical concepts, such as uncontrolled diabetes, into machine-readable formats, agentic AI can understand patient eligibility and assist in recruitment, but humans still make the final decisions.

“The human remains in control. The agentic AI acts as an augmentation layer — like a junior data manager or an expert assistant — improving efficiency without replacing judgment,” Rao says.

Looking ahead: agents designing trials

Rao envisions a future where agentic AI does more than manage data — it could design entire clinical trials. Agents could model patient cohorts, simulate outcomes, optimize trial arms, and even anticipate regulatory considerations. Multi-agent orchestration could break down silos between statisticians, clinicians, and regulatory teams, allowing virtual collaboration at scale.

“Agents can simulate hundreds of trials, model comorbidities, and predict outcomes before a single patient is enrolled. That could dramatically increase the probability of trial success,” Rao says.

The implications for oncology are particularly striking. Molecule-to-market timelines currently take eight years; Rao believes agentic AI could shorten this to four or five years, bringing life-saving treatments to patients faster.

“If molecule-to-market takes eight years today, agentic AI could help bring that down to four or five years. Patients are dying — everyday counts,” he says.

Advice for pharma leaders

Rao offers clear advice for pharma executives: don’t let technology drive adoption. Start with science, identify real pain points, and use agentic AI to address them.

“Technology is the bottom of the pyramid. The top is science. Identify pain points in processes, change management, or data ecosystems. Use AI as an enabler, not a hammer looking for a nail,” he says.

Meaningful adoption requires embedding agentic AI into workflows, so users see real value. Solutions must solve tangible problems, demonstrate efficiencies, and gain user trust, otherwise adoption stalls.

“I’ve seen enterprise AI rolled out and ignored because it didn’t align with meaningful daily tasks. The same is true for agentic AI — start with use cases that make a difference and build from there,” Rao adds.

Standing on the shoulders of giants

Rao concludes with a metaphor that captures his vision for agentic AI:

“Isaac Newton said he could see so far because he stood on the shoulders of giants. In my mind, a large language model is a giant — it has read billions of pages and compressed that knowledge. Why not stand on its shoulders to see further and save human lives?”

For Rao, agentic AI is a giant that augments human expertise, accelerates decision-making, and opens the possibility of radically faster drug development, without ever replacing human judgment.