Agentic AI in Action: Real-world lessons from Cedars-Sinai, Dayton Children's and Epic

Over the past year, health systems have eagerly embraced ambient AI assistants—helping clinicians feel more supported, cutting down on documentation time, and making patient interactions more human and personal. Now, with all this progress and growing interest in bringing AI even deeper into clinical workflows, a new concept is stepping into the spotlight: Agentic AI.

On May 15, 2025, Becker’s Hospital Review hosted a webinar titled "Agentic AI in the EHR: From Ambient Notes to Autonomous Agents." This session brought together industry leaders to explore how agentic AI is transforming electronic health record (EHR) workflows and redefining clinical care. The webinar featured:

• Dr. Shaun Miller, MD, MBA – Chief Health Informatics Officer at Cedars-Sinai
• J.D. Whitlock – Chief Informatics Officer at Dayton Children’s Hospital
• Garrett Adams – Vice-President of Research & Development at Epic Systems

Curious how healthcare leaders are thinking about what’s next? Here are some of the ideas that stood out to us—and felt worth sharing.

Setting the Stage: Understanding Agentic AI

But first, what exactly is agentic AI? Moderator Jakob Emerson, Associate News Director at Becker’s Hospital Review, set the stage by exploring the concept. Garrett Adams defined it simply but powerfully: "Agentic AI doesn’t just assist—it takes initiative. It’s like having a digital teammate that can understand, decide, and act." This new breed of AI is capable of not only observing and documenting but also analyzing data, making decisions, and even performing complex tasks autonomously.

Agents go beyond prompt-response. They have autonomy, goals, memory, planning capabilities, and can use tools or APIs to take action. "Agents represent the possibility of a digital workforce... that may rebalance the care team-to-patient ratio," Adams explained.

Real-World Implementations: Transforming Clinical Workflows

The speakers shared how their organizations are leveraging early agentic AI capabilities to improve EHR workflows, support clinician decision-making, and streamline administrative tasks.

Dr. Shaun Miller discussed how Cedars-Sinai has gone beyond traditional ambient AI. The organization has deployed over 100 use cases—about one-third already in production. These include predictive algorithms for maternal health and sudden cardiac death, early adoption of imaging AI through Aidoc, and the use of Nabla for ambient documentation in outpatient care. “It’s been life-changing in terms of saving time on notes and just pajama time outside of the clinic,” Miller said. Cedars-Sinai also uses AIVA for nursing documentation and is building backend models in its Department of Computational Biomedicine.

J.D. Whitlock shared that Dayton Children’s Hospital focuses on both predictive and generative AI. Predictive tools include sepsis alerts and remission risk predictors. The hospital is midway through its ambient AI implementation and preparing to deploy pre-visit agents that help flag missing labs, set expectations, and streamline handoffs. “Where do patients arrive confused, and how can we solve that with an agent? That’s where we started,” Whitlock noted.

Garrett Adams highlighted how Epic is embedding generative AI directly into clinician workflows. Examples include patient summaries, discharge planning, and nurse shift handoffs. Instead of external chatbots, Epic embeds purpose-built agents within its ecosystem. "It’s intuitive, and they don’t need to learn new systems—it’s where they already are," Adams emphasized. Adams also highlighted Sidekick, a conversational layer on top of SlicerDicer that reduces barriers to self-serve analytics.

Overcoming Challenges: Integration, Simplicity and Collaboration

Despite the clear benefits of agentic AI, the panelists acknowledged several challenges. Dr. Miller emphasized the importance of aligning AI with clinical workflows. "If it’s not intuitive, it won’t work—no matter how advanced it is," he said.

As healthcare complexity increases, Dr. Miller expressed hope that agentic AI will restore simplicity. “It’s becoming increasingly complex. The opportunity here is to make it simple again—everyone speaking the same language,” he said. This includes AI helping coordinate backend tasks such as appointment scheduling, referral routing, and pulling relevant data—all through natural language interfaces.

Multimodal Intelligence for Better Patient Support

Dr. Miller spoke about the growing potential of combining various data types—voice, chart, physiologic signals, and patient questionnaires—to enable real-time clinical decision support. "Taking that step further to help the patient almost more directly orchestrate the care... that’s where patients are seeing value," he shared.

This vision includes AI not only supporting physicians during visits, but also helping patients manage follow-up tasks, from scheduling to referrals.

Building Guardrails: Safe and Disciplined Implementation

Shaun Miller emphasized the need to proceed responsibly. "Let’s make sure we do it safely... find the impactful areas, but still focus on lower-risk use cases first," he said. Cedars-Sinai relies heavily on existing clinical protocols, validated datasets, and structured oversight to ensure that AI augments care without introducing risk.

J.D Whitlock reinforced the need for strict data privacy and lightweight implementations suitable for smaller health systems. “We have a responsibility to make sure our youngest patients are protected,” he said, pointing to pediatric-specific challenges.

Garrett Adams reiterated the importance of embedding governance into development from day one. "Governance is not an afterthought—it is a core pillar of our AI strategy," he explained.

AI Supervising AI

Looking ahead, Miller also addressed the layered nature of AI governance. "We need an AI model to supervise AI… the holy grail is not having to always be the one saying: ‘No, don’t give me that answer.’" This concept of meta-AI—models that monitor, evaluate, and refine the behavior of other models—could be crucial to future scalability.

That’s a Wrap, For Now

Agentic AI is starting to show up in real, practical ways across healthcare. As Dr. Miller stated, "We’re just scratching the surface of what agentic AI can achieve."

For those who missed the live session, the full recording is available through here.