This talk highlights practical ways agentic AI can reduce routine manual work across clinical development. Emphasis is placed on what is dependable today, how to implement it safely, and where human oversight remains essential.

Recent breakthroughs have enabled LLMs to carry out multi-step scientific tasks through sophisticated tool use, reasoning, and orchestration. We develop an agentic AI framework to assist with decision making on clinical programs. This framework consists of multiple agents which integrate evidence from experimental data and scientific text to address complexity of the modern clinical programs. This framework demonstrated promising results while applied to several clinical programs.

Getting the right therapy to the right patient at the right time requires a patient-centric evidence approach. We share advanced modeling methods that enable enriched patient understanding using structured and unstructured medical information. We apply these techniques on topics like patient journey to identify predictors of loss of control. We combine these insights with patient panels, literature, physicians, and health technology assessment input to shape disease strategy and accelerate protocol and development planning decisions. Lastly, we show the promise of these approaches beyond development into medical to support launched products.

We have designed and implemented a clinical-trial forecasting system powered by ML models and heterogeneous data using software, modeling, and systems best practices. Our flexible and scalable architecture allows us to rapidly iterate our models while maintaining a clear line of sight to source data and results. If Biogen can build this ourselves, other pharmaceutical companies can too.

We are transforming the way we are running clinical trials leveraging a new concept: agentic AI. Our new AI-powered assistant will transform analytics from passive dashboards into proactive insights and actions, supporting clinical teams to be aware of key activities they need to take, organize their day and even take some actions for them.

We’ll share use cases of how we implemented an operating model with data-driven decision-making at its core. Leveraging the availability of more data than ever, we changed our ways of working to make the most out of it.

Purpose: To automate and expedite the creation of the Lay Protocol Synopsis (LPS) and Brief Summary (BS), which are time-sensitive regulatory documents needed for effective communication with ethics committees, patients, and the public. Method: Utilizes a library of prompts mapped to the CTP, along with a blend of rule-based template language, large-language models (LLMs), and retrieval augmented generation (RAG) to produce an initial draft of the LPS.

The “perfect” clinical operations vendor platform does not exist and probably never will. Even leading suites cover only ~60–80% of real-world needs once governance steps, audit evidence, edge cases, and post-acquisition system realities enter the room. Organizations typically respond by force-fitting the platform, and end up paying twice: Process debt (manual workarounds, shadow processes, spreadsheet sprawl), and integration debt (fragile point-to-point connections, duplicated data, inconsistent metrics). In this session, we propose a different operating model: Core Platforms + Niche Capability Vendors + a thin, reusable Microservices Layer. The outcome: fewer workarounds, more consistent data, and teams spending time on decisions that improve trial delivery and ultimately patient impact.