AI is rapidly changing clinical infrastructure across BJC HealthCare and the WashU School of Medicine. A recent presentation explained how the behind-the-scenes work of IT teams is making these changes happen seamlessly, reliably and securely.
During the March IT All Hands meeting, Richard Biehl and Michael Kriemelman, Senior Directors of Application and Product Delivery, delivered an exciting presentation on how a growing suite of AI-enabled tools embedded directly into Epic software is reducing documentation burden, accelerating patient communication and reshaping how clinicians interact with technology. Kriemelman and Biehl outlined how ambient documentation, generative AI and machine-learning capabilities are now operating at enterprise scale, tightly integrated with Epic workflows and supported by foundational IT systems.
“None of this works without the infrastructure, networking, security and device teams,” Biehl said. “Even if you’re not on the Epic team, you’re part of the Epic team. This is truly full-stack clinical innovation.”
Ambient AI documentation at scale
At the center of the discussion was the Abridge partnership, an ambient AI solution integrated into Epic that uses mobile devices to capture patient-provider conversations and generate structured clinical documentation in real time.
From a technical perspective, Abridge functions as an ambient listening layer within the Epic ecosystem. Using device microphones approved for clinical use, the platform differentiates between patient and clinician voices, applies specialty-specific language models, and generates draft clinical notes aligned with Epic note templates—before pushing the structured output back into the electronic health record (EHR).
The platform currently supports more than 50 specialties, 28 languages and multiple care environments, including ambulatory settings with pilots underway for inpatient and emergency departments. Its expansion has required close coordination across mobile device management, wireless networks, identity access, Epic integration points and data security.
“This is a deeply embedded workflow tool, not a bolt-on,” Biehl said. “It listens during the visit, pulls in chart context, applies specialty logic and produces notes that are ready for clinical review.”
Since its broader rollout, more than 800,000 notes have been generated across BJC and WashU, with Abridge reporting an 85% content retention from the Abridge draft correlating to a reduction in clinician documentation effort. Adoption continues to rise, with thousands of notes now generated daily and near-universal user retention.
Behind the scenes, the effort represents a long-term, multimillion-dollar innovation agreement between BJC, WashU and Abridge—designed not only for current note generation, but for future-state workflows such as revenue cycle support, specialty documentation enhancements and inpatient expansion.
Reducing message burden with generative AI
Kriemelman shifted the focus to another high-volume workflow: secure messaging through MyChart.
Across the enterprise, clinicians now receive more than 5 million patient messages annually, driven largely by telehealth expansion and increased portal adoption. For IT and clinical informatics leaders, the challenge has been finding ways to scale communication without scaling burnout.
That challenge led BJC and WashU to become early adopters of Epic’s Augmented Response Technology, or ART—an AI-driven message-drafting capability embedded directly into Epic In Basket workflows.
“ART reads the patient message, looks at relevant chart data and drafts a response—but the clinician is always the final decision maker,” Kriemelman said. “Nothing goes out without review. That governance model was non-negotiable.”
From a technical standpoint, ART relies on generative AI models governed by Epic’s architecture, layered with extensive local configuration. Informatics teams worked through thousands of prompts, responses and validations to ensure safety, tone consistency and clinical appropriateness—while monitoring for hallucinations and edge cases common to large language models.
The system generates draft responses, which clinicians can edit, personalize or discard. Adoption metrics are tracked directly within Epic, including thumbs-up and thumbs-down feedback loops that inform iterative model tuning.
Since its full ambulatory rollout earlier this year, ART has generated 40,000 to 50,000 draft responses monthly, supporting message turnaround times that now average less than half a day for 95% of patient messages.
“The technical win here isn’t just AI,” Kriemelman said. “It’s how tightly it’s integrated into existing Epic workflows, so clinicians don’t have to change how they work.”
IT infrastructure as an enabler
Both leaders emphasized that success depends on enterprise IT readiness. Ambient recording, real-time transcription and generative AI require resilient wireless networks, secured endpoints, identity management, device lifecycle governance and continuous monitoring.
“This kind of AI is unforgiving if your foundation isn’t solid,” Biehl said. “Latency, dropped connections, device issues—those become clinical problems very quickly.”
The Epic-1 team continues to work closely with infrastructure, security and operations teams as additional AI capabilities move from planning to piloting and production, including inpatient summaries, AI-assisted nursing documentation, revenue cycle automation and analytics enhancements.
Some tools have already been tested, paused and reworked—underscoring a deliberate, feedback-driven approach. “If it doesn’t deliver value or users aren’t satisfied, we stop and reset,” Biehl said.
A scalable model for clinical innovation
Together, Abridge and ART offer a model for how health systems can operationalize AI: tightly integrated, clinician-governed, metrics-driven and supported by enterprise-grade IT platforms.
“We’re not trying to replace clinicians’ judgment,” Kriemelman said. “We’re using technology to take the administrative weight off so they can practice medicine—and that only works when IT and clinical teams are aligned.”
As Epic’s AI roadmap continues to expand, leaders said the underlying formula will remain the same: strong infrastructure, thoughtful governance and technology that works quietly in the background—so clinicians can stay focused on patients.