It’s been said that the only constant in the world is change.
For decades, clinical trials have been a human only endeavor, with teams of clinicians, study teams, and patients working hand in hand to bring the latest molecules to market. Now, a new central actor has entered the clinical paradigm, agentic artificial intelligence.
Only three years after OpenAI kicked off the artificial intelligence arms race, AI has gone from requiring users to prompt it, to pre-emptively identifying bottlenecks, safety risks, and more, thanks to agentic AI.
Agentic AI is an autonomous, goal-oriented system that uses reasoning and external tools to independently plan, execute, and adapt multi-step actions with minimal human intervention to achieve complex objectives.
Sponsors and CROs have begun using AI agents across their workforces to improve trials in ways that humans have traditionally struggled to accomplish. For instance, organizations have been creating AI agents to analyze prior trial protocols, benefiting from lessons learned across prior trials and real-world outcomes, enabling teams to anticipate risks and automating elements of submission drafting. Anomaly detection has helped teams better identify outliers in operational metrics or safety signals, prompting early interventions. Document intelligence accelerates medical writing by grounding generative outputs in verified data, which reduces cycle time without sacrificing accuracy.
However, in each of these use cases, humans remain squarely “in the loop.” Or rather, the decision making isn’t left entirely to AI. Instead, the objective is to augment clinical, regulatory, and legal teams with tools that surface the right information at the right time.
This core concept, keeping a human in the loop, is essential to clinical decision making and operations as agentic AIs, while powerful, are not inherently suited to fully autonomous operation in all regulated contexts.
Clinicians leveraging agentic AI for structured and repeatable tasks
Agentic AI excels at processing large volumes of structured information and performing routine actions at speed. This capability is particularly valuable in clinical research, where data is often distributed across numerous systems and requires significant manual effort to aggregate and analyze. AI agents that have direct system access can complete these tasks within seconds.
Agents can handle routine administrative work by following up with investigational sites about missing documents, monitoring data entry timelines, tracking protocol deviations, generating reports, managing correspondence, and coordinating meeting schedules.
They can also support data processing by performing initial cleaning and validation, producing statistical summaries, preparing draft enrollment reports, running database queries, and assembling standardized safety data. By automating these activities, agentic AI allows clinical teams to shift their attention to higher value work that requires strategic thinking and clinical judgment.
Areas where human oversight remains essential
Despite these efficiencies, many mission-critical tasks in clinical development require specialized human judgment that AI cannot reliably provide. Regulatory bodies expect, and in many cases mandate, that humans review activities affecting patient safety and trial integrity.
Serious adverse event reporting depends on clinical interpretation to understand safety signals and ensure accurate and timely communication to regulators. Regulatory filings such as INDs and CTDs require careful human review to confirm that documents are complete, accurate, and consistent with current standards. Any protocol amendment must be evaluated by qualified medical and regulatory professionals who can assess its potential impact on patient safety and study outcomes.
Human oversight is also vital in clinical data management. Although AI can identify discrepancies, humans decide which issues are clinically meaningful and how they should be prioritized. For instance, medical coding requires judgment that depends on clinical expertise. Final data lock decisions must involve human review to ensure the accuracy and integrity of the clinical database. Quality assurance and compliance activities further emphasize the need for human involvement. Regulatory audits require human judgment, contextual understanding, and direct interaction with inspectors. Risk assessments rely on experience and intuition that go beyond pattern recognition. Corrective and preventive action plans must be designed with a deep understanding of root causes and organizational dynamics.
Designing an effective human and AI partnership
Organizations that achieve meaningful progress with agentic AI do so by establishing strong governance frameworks that define how humans and AI should work together. Clear boundaries indicate what AI may independently perform and where human judgment is required. Structured approval workflows ensure that regulated outputs always receive appropriate review.
For example, escalation procedures help teams manage ambiguous or high-risk situations. Role-based oversight ensures that senior reviewers are involved in the most critical activities, while routine tasks may be delegated to junior staff supported by AI. Ongoing training helps supervisors remain effective evaluators of AI outputs and equips them to identify errors or limitations.
This balanced approach not only preserves regulatory compliance but also creates a stable environment for organizational learning. As teams deepen their understanding of AI capabilities, they can gradually expand the level of autonomy these systems are allowed to take on.
Building responsible practices through human-guided AI
Agentic AI can significantly accelerate the pace of clinical development when humans remain in control of the most sensitive and high-impact decisions. When deployed thoughtfully, these systems support continuous progress by completing routine activities at speed while humans apply their expertise to tasks that demand nuanced understanding and contextual reasoning. Organizations that adopt a responsible human in the loop approach today will not only improve current development timelines but will also lay the foundation for a more efficient and intelligent clinical ecosystem in the future.
A recent article by Hoag Levins, the Editor of Digital Publications at the University of Penn, Leonard Davis Institute of Health Economics, emphasizes that implementing AI in healthcare is not about a technological miracle but about years of deliberate, careful work.
Systems need to be designed with transparent interactions, clinician customizability, safety, trust, and full accounting of medical, legal, regulatory, and reimbursement implications. The takeaway is that healthcare organizations must intentionally build toward a “human on the loop” future rather than assume mere addition of AI will yield better outcomes.
Human-guided intelligence creates new ways of working
As clinical development enters this new era, the opportunity is not to replace human expertise but to amplify it. Agentic AI offers unprecedented speed, scale, and insight, yet its true value emerges only when guided by the judgment, ethics, and experience of the people who safeguard patient wellbeing. The future of clinical trials will be shaped by organizations that recognize this balance, invest in deliberate design, and commit to responsible oversight. By keeping humans in the loop while embracing the strengths of advanced AI, the industry can achieve safer, smarter, and more efficient research that ultimately brings better therapies to patients faster.
.webp)
