In the modern clinical trial landscape, the sheer volume of data has transcended human capacity. Between 2020 and 2025, the average Phase 3 clinical protocol saw its data collection surge from approximately 3.56 million data points to 5.96 million—a staggering 67% increase in just five years. This explosion of information is 6.4 times higher than the 2012 average of roughly 929,000 data points, according to collaborative research from TransCelerate BioPharma and the Tufts Center for the Study of Drug Development.
As the industry pivots toward artificial intelligence (AI) to manage this deluge, a new paradigm is emerging: the autonomous clinical trial. However, as the technological capabilities of AI and machine learning (ML) models grow, the industry is discovering that the most critical job in a clinical trial isn’t just generating data—it is the human-led supervision of the agents that process it.
The Data Deluge and the AI Disincentive
The TransCelerate and Tufts study highlighted a concerning trend: the ease with which AI can process data is acting as a "disincentive to reduce data volume." Researchers found that nearly one-third of Phase 3 procedures and their associated data points were classified as non-core or non-essential.
"I don’t think sponsors are as worried about collecting too much data anymore," says Venu Mallarapu, chief transformation and AI officer at eClinical Solutions. Mallarapu argues that while AI has made these massive datasets manageable, it has simultaneously created a "data-hoarding" mentality. Sensors, wearables, and continuous monitoring tools are pushing volumes to new heights, and while AI makes the crunching of that data significantly easier, the fundamental question remains: are we collecting the right data?
Chronology: From Pilot Programs to Agentic Workflows
The adoption of AI in clinical trials has followed a rapid, phased trajectory over the last several years:
- 2020–2022 (The Foundation): The industry focused on digitizing records and utilizing basic ML for data cleaning. AI was largely seen as a backend tool for optimizing existing manual workflows.
- 2023–2024 (The Rise of Assistance): AI began to appear in the form of "assistive" tools. Clinical research associates (CRAs) started using AI to draft reports or triage patient materials, with human intervention required at every step.
- 2025–2026 (The Agentic Era): The industry shifted toward "scoped agents"—AI programs capable of executing tasks within a reengineered workflow. Rather than just assisting, these agents now perform end-to-end tasks like SDTM mapping, with humans acting as supervisors who review, approve, or reject the output.
This timeline reflects a broader shift toward enterprise-wide AI implementation. A recent poll of 300 life sciences professionals by the Pistoia Alliance revealed that 30% of organizations have reached enterprise-wide AI status, with value concentrated heavily in regulatory and reporting documentation.
Supporting Data: Efficiency vs. Reality
The integration of AI into clinical trials is not without its hurdles. An Everest Group survey of 200 senior pharma, biotech, and CRO decision-makers commissioned by Medidata found that 82% of organizations using AI in trial operations have 18 months or less of experience.
Despite the hype, the industry is cautious: 63% of these organizations explicitly mandate human oversight. The most successful applications of AI—often reporting results that exceed expectations—remain in bounded, low-risk areas:
- Task and workflow automation: 46.5% reported high success.
- Data cleaning: 40.5% reported high success.
- Query resolution: 36.5% reported high success.
The discrepancy between the potential for "autonomous" trials and current reality is stark. As Dr. Pamela Tenaerts, Chief Medical Officer at Medable, points out, "I would argue protocol optimization still needs to happen. An agent may be able to help a monitor navigate millions of data points, yet its processing capacity leaves the scientific justification for collecting them untouched."
Official Responses and Expert Perspectives
The role of the human is being redefined from "doer" to "supervisor." Industry leaders emphasize that the "agent proposes, human disposes" model is the gold standard for regulatory compliance.
The Role of Human Oversight
"No agent directly takes an action without human approval," explains Mallarapu. "We log everything: what triggered the agent, who it acted on behalf of, the version of the data, the program used, and the human decision with its timestamp. This provides the audit trail, traceability, and transparency required by regulators."
This supervision is essential because, as seen in recent experiments with large-scale agentic models (such as the METR investigation into AI cybersecurity evaluations), agents left to coordinate without sufficient oversight can become "unreliable." When 1,200 agent instances attempted to coordinate a task, they generated thousands of pages of output, often missing critical findings and making errors that only a human could detect.
"If you heavily delegate analysis to AI agents, you often fail to highlight the most important findings," the METR report concluded. "A human researcher, given enough time, would likely have produced more calibrated and useful analysis."
Managing the "Balloon" Effect
Dr. Tenaerts notes that automation creates a "push-the-balloon" effect. "If you become dramatically more efficient at sending queries and emails to sites, the sites simply absorb the load," she says. "If you dump all that stuff on them, that’s still a bottleneck. You need to figure out the whole system, not just the part you’re automating."
Implications for the Future of Clinical Research
The shift toward autonomous trials has profound implications for the clinical trial ecosystem, from preclinical research to site management.
Re-Engineering for Agents
"If you are introducing a system or automation, you need to think about re-engineering the process as well," says Mallarapu. This involves centralizing data architectures using platforms like Snowflake or Databricks. By moving away from fragmented spreadsheets toward a unified data lakehouse, companies can provide agents with a coherent view of information while maintaining the ability to extract data for human review.
The "Swarm" of Agents
In the near future, we will see "swarms" of agents working in concert. eClinical’s Data Advisor already functions this way: a user interacts with one interface, while a swarm of agents manages data mapping, identifies discrepancies, and drafts responses behind the scenes. This allows the human expert to maintain a global view of the trial rather than being buried in 13 different tabs on a computer screen.
The Human-in-the-Loop Imperative
As the industry moves toward the vision of a "self-driving" clinical trial, the human-in-the-loop remains the cornerstone of safety and efficacy. Whether it is an agent flagging a new medication that contradicts a safety report or an AI agent mapping complex clinical data, the final validation remains the prerogative of the human.
"It’s somewhat analogous to what’s happening on the internet today," says Ken Getz, Executive Director of the Tufts Center for the Study of Drug Development. "Very few people are actually visiting websites; they’re relying on AI engines to interact with AI optimization. It’s fascinating where this may ultimately lead."
Conclusion: Balancing Innovation with Integrity
The dream of the autonomous clinical trial is accelerating, driven by the need to manage massive datasets and the desire to reduce burnout among site staff. However, the current reality is one of "assisted" autonomy.
For sponsors, the challenge is not just to integrate AI, but to do so with the discipline to strip away non-essential data collection. As regulators continue to emphasize that critical-to-quality factors must remain "uncluttered," the human supervisor will remain the final arbiter of trial integrity. In the age of the agent, the most valuable skill in clinical research is no longer data collection—it is the ability to oversee, audit, and interpret the work of the machines that have become our most powerful partners.
The future of clinical trials will be defined not by how much data we can collect, but by how effectively we can ensure that our automated agents act in service of clear, scientific, and patient-centered goals. The balloon of complexity will continue to shift, but with the right architecture and human-led oversight, the industry is poised for a new era of efficiency.
About the Author
