For decades, running a clinical trial has meant layering manual processes on top of digital systems. A medical center enrolls patients and records their data in electronic health records. Staff then re-enter that data by hand into a separate data-capture system. The drug company developing the medicine reviews the re-entered data and submits it to the FDA. The FDA reviews it again.

Every handoff in that chain introduces delay, transcription error, and cost. The process has not changed meaningfully in the time electronic health records have existed. The data was already digital. The re-entry was the bottleneck.

On Monday, the FDA launched a pilot program to eliminate that bottleneck. AI and data science tools will extract clinical trial data directly from electronic health records and transmit it to the agency and the sponsoring pharmaceutical company in real time. No manual re-entry. No months-long lag between data collection and regulatory review.

Dr. Emma Meagher, senior vice dean for clinical and translational research at Penn Medicine, framed the status quo plainly: "In general, we have a sense that the way we do clinical trials is dysfunctional in many ways."

The Pilot Is Already Running

This is not a proposal. The FDA announced two proof-of-concept clinical trials that are already reporting endpoints and data signals to the agency in real time. AstraZeneca and health technology firm Paradigm Health have validated safety data from a midstage trial testing a drug for mantle cell lymphoma in previously untreated patients. The data moved through the system successfully.

The agency has also released a Request for Information for a broader pilot program expected to launch this summer. The stated ambition is significant: the FDA believes this approach could cut years from drug approval timelines and help the United States compete with China in biotechnology.

The competitive framing is worth noting. The FDA did not lead with patient outcomes or cost savings. It led with national competitiveness. That tells you what actually moved the decision.

"The data was already digital. The re-entry was the bottleneck. AI did not create new information here. It eliminated the manual step between where the information already lived and where it needed to go."

Generative AI Gets the Headlines. Analytical AI Gets the Results.

The current AI conversation is dominated by generation: images, code, marketing copy, entire applications from a prompt. Those capabilities are real and commercially significant. But the FDA pilot is a reminder that the highest-value AI deployments in regulated, data-heavy industries look nothing like that.

This pilot uses AI to read structured and unstructured data from electronic health records, validate it against trial protocols, flag anomalies, and transmit it in a format regulators can act on immediately. There is no creative output. There is no chatbot. There is a system that does in seconds what used to take weeks of manual transcription and review cycles.

I wrote about a related dynamic earlier this week with Isomorphic Labs, where the question is whether AI-designed drug candidates survive contact with human biology. The FDA pilot sits on the other side of that problem. It is not about designing the drug. It is about getting the trial data to the people who approve the drug without losing months to manual processes.

Both matter. But the FDA pilot is closer to production. It is already running. And it addresses a problem that every technology leader in a regulated industry will recognize: two systems that hold the same data, connected by a person with a keyboard.

The Pattern Enterprise Technology Leaders Should Recognize

Clinical trials are not the only place where humans serve as the integration layer between systems that should talk to each other. The same pattern exists in financial services, where trade data moves between front-office and back-office systems through manual reconciliation. It exists in insurance, where claims data is re-keyed from intake forms into adjudication platforms. It exists in manufacturing, where quality inspection data recorded on the floor gets manually entered into ERP systems hours or days later.

In every case, the data is already digital at the point of origin. The re-entry exists because the systems were built at different times, by different vendors, with different data models, and nobody built the bridge. AI is now building those bridges faster than integration middleware ever did, because it can handle the messiness — inconsistent field names, unstructured notes, missing values — that traditional system integration cannot.

The FDA pilot is a proof point for that broader pattern. If a federal regulatory agency with some of the most stringent data requirements in the world is willing to accept AI-extracted data from electronic health records in real time, the bar for enterprise adoption just moved.

The Uncomfortable Timeline Question

If AI can extract and validate clinical trial data in real time today, the obvious question is why anyone was manually re-typing it yesterday. The answer is institutional, not technical. The process existed before the technology caught up, and no single stakeholder — not the medical centers, not the drug companies, not the FDA — had sufficient incentive to rebuild it unilaterally.

What changed is external pressure. The FDA explicitly cited competition with other nations as a driver. When the agency frames a data infrastructure pilot as a matter of national competitiveness in biotechnology, the internal calculus shifts. Legacy workflows do not die because someone invents a better way. They die because the cost of not changing becomes visible to the people who control budgets.

The FDA's Real-Time Clinical Trials pilot program RFI is open now, with the broader pilot expected to launch summer 2026.