Alex Bedenkov, VP, Global Evidence, BioPharmaceuticals Medical, AstraZeneca.
We do not conduct studies just to publish; we do so to generate practice‑changing evidence that improves patient outcomes and lives. Yet the journey from study design concept (SDC) to clinical guideline citation is often fragmented by handoffs, delays and lost momentum.
Imagine a seamlessly orchestrated flow—studies designed smarter, delivered more smoothly, risks reduced and communications faster and more transparent from SDC to publication. Agentic AI is beginning to unlock this future for evidence generation, especially in Phase 3b/4 trials.
From Siloed Steps To A Seamless Ecosystem
The breakthrough is not a single tool but an ecosystem of AI agents whose outputs become one another’s inputs—each auditable and grounded in real‑world data (RWD) and evidence. The effect is fewer handoffs and blind spots and more human time where it matters.
• Design and protocol agents co‑design smarter studies by mining prior protocols, scientific literature and regulatory context, creating control arms and stress‑testing inclusion/exclusion criteria and endpoints against real‑world prevalence and operational feasibility.
• Authoring agents partner with subject matter experts to transform structured specifications and available data into first‑pass drafts for SDCs, protocols, case report forms/informed consent forms, clinical study reports (CSRs) and manuscripts—preserving traceability and change history so humans focus on interpretation rather than manual drafting or reformatting.
• Site and investigator selection agents integrate location‑relevant RWD and historical site performance to identify optimal trial partners, reducing amendments and accelerating activation and enrollment.
• Study‑delivery agents operate where GxP, patient safety and external scrutiny converge—fusing digital biomarkers and EHRs for near‑real‑time safety signal detection, prioritizing source data verification and operational drift monitoring and supporting sites and patients with assistants that coordinate reminders, logistics and patient‑reported outcomes quality checks.
• Biostats and data agents automate reconciliation and cleaning and generate programmed outputs to speed database lock.
RWD As A Strategic Amplifier
Most late‑stage pain in evidence generation is “designed in” early. Grounding study designs in RWD, care‑pathway realities and site capacity before protocol freeze reduces amendment risk and makes recruitment plans credible. When RWD is treated as a strategic asset—and embedded in an enterprise knowledge layer that unifies prior studies, medical literature, operating experience and regulatory texts—teams can detect diagnostic signals earlier, standardize care where appropriate and evaluate pathways at scale.
Location‑relevant RWD shifts trial design from reactive, hospital‑centric models to proactive, patient‑centric approaches aligned with real care patterns and unmet needs. Ultimately, it’s not only execution excellence but RWD‑driven, patient‑centered design that shapes timelines, data quality and post‑readout relevance—provided data are used ethically, with explicit privacy protections and appropriate governance.
Closing The AI Blind Spot In Delivery
Today’s visible AI tools cluster around lower‑risk, narrower steps such as design and authoring. The externally facing, GxP‑sensitive delivery phases—real‑time safety oversight, site monitoring and patient support—remain under‑addressed relative to their potential, as they demand advanced capabilities, robust validation, traceability and human‑in‑the‑loop governance. Yet this is precisely where integrated, well‑governed AI can drive step‑change gains in speed, quality and predictability.
We should also resist “demo theater.” Value arrives when agents are grounded in high‑quality data, governed for safety and embedded in accountable workflows—not when a new model reaches a showcase. Any AI used in evidence generation must respect patient privacy, copyrights and appropriate data use and storage. The path forward is to engage these high‑value zones with the right assurance disciplines—defined autonomy levels, model and agent change control and continuous performance monitoring with clear escalation to clinical experts.
Six Priorities For Building An Impactful Agentic AI Ecosystem
1. Align with outcomes that matter to patients and clinicians.
Anchor AI to hard business outcomes—on‑time site activation, fewer amendments, cleaner data at database lock, faster CSR and publication velocity—and measure the deltas with the same rigor applied to trial endpoints.
2. Treat data as a strategic product, not an export.
Agentic AI without governed data products is theater. Build a “clinical knowledge foundation” that unifies protocols, outcomes, site performance, RWD and regulatory texts with lineage, standards, access policies and privacy‑by‑design (e.g., minimization, federated approaches) to enable safe reuse.
3. Solve real problems first.
Prioritize workflows that unblock timelines or reduce risk—safety triage, monitoring and data quality—over “nice‑to‑have” experiments. Integration and change management, not model choice, determine value at scale.
4. Build trustworthy, human‑centric, responsible AI.
Involve workflow owners and subject‑matter experts from day one. Ground outputs with citations. Log decisions, design for explainability, and institute bias assessment, human oversight and transparent consent—especially where patient data and clinical decisions intersect.
5. Avoid the model trap; engineer the system.
Models are a small part of the effort. Integration, orchestration, observability, guardrails and change management are the rest. Treat agents as products, standardize policies and runtime controls, and manage economics—run costs can exceed build costs at scale.
6. Operationalize governance, don’t just cite it.
Translate responsible AI and multi‑stakeholder frameworks into operating routines: autonomy levels, validation and change control, continuous performance monitoring and escalation to clinical experts.
The Time To Lead Is Now
Agentic AI can accelerate timelines, improve data quality and make trials more inclusive and efficient. The greater impact lies in extending these gains into safety oversight, monitoring and patient support—with assurance mechanisms that sustain trust.
Take decisive steps:
• Move beyond pilots and point tools. Build an ecosystem where agents work in harmony across the life cycle, not in silos.
• Make business ownership and integration nonnegotiable. Treat agentic AI as a catalyst for operating‑model change.
• Prioritize trust, transparency and human partnership. Maintain multi-stakeholder dialogue, document provenance, and keep humans in the loop where safety and ethics demand it.
We owe patients speed with standards, ambition with accountability and innovation with integrity. The opportunity is here—if we choose to build it, end‑to‑end, the right way.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?