Brown University Health is scaling Microsoft Dragon Copilot across its emergency departments, introducing ambient artificial intelligence that automatically documents patient encounters as they happen. The deployment, detailed in a new Microsoft customer story, targets one of healthcare’s most persistent friction points: the clerical load that forces clinicians to split their attention between keyboards and the people in front of them.

What’s Changing at Brown Health

For physicians like Dr. Anthony Napoli, executive vice chair of emergency medicine, Dragon Copilot already marks a shift. During a recent shift at Rhode Island Hospital—a Level I trauma center that handles hundreds of thousands of visits annually—Napoli used the tool while speaking with an elderly patient reporting dizziness and palpitations. The ambient system captured the conversation despite hallway noise, letting him maintain eye contact and focus on clinical reasoning while the AI produced a structured draft note.

That draft, not a final record, is the core deliverable. Dragon Copilot listens to an authorized patient encounter, transcribes relevant portions, and generates a clinical summary for the provider to review, edit, and sign. It is not a replacement for professional judgment, and it does not independently enter information into the electronic health record. Instead, it shifts documentation from a synchronous task competing for attention during visits to a post-encounter validation step.

Brown Health’s rollout goes beyond single-purpose transcription. The organization is also piloting AI agents—workflow assistants that can summarize records, draft patient communications, and route administrative tasks. Together, these moves reflect an enterprise strategy to embed AI across clinical and operational workflows, not just in isolated note-taking.

What This Means for Healthcare Teams

For Clinicians

The most immediate benefit is a reduction in after-hours charting. Physicians often face a grueling choice: type during visits and risk weakened rapport, jot hurried notes and transcribe later, or stay late to finish documentation. Ambient AI changes the sequence: the system does the initial organization, and the clinician confirms accuracy. At Brown Health, the goal is to give back time for patient interaction and clinical decision-making.

But that gain comes with a responsibility. Drafts generated by large language models can be fluently wrong. Automation bias—the tendency to trust polished text—could let errors slip into the permanent record. Every note must be verified. Brown Health’s leadership stresses that drafts are starting points, not finished products.

For IT Administrators

Implementing Dragon Copilot isn’t a simple plug-and-play exercise. The platform supports web, desktop, and mobile access, and it can be integrated directly into compatible EHR workflows. Integration quality determines success. Standalone use—where clinicians copy and paste from a Dragon-generated draft—introduces manual transfer risks. Embedded deployment, where the AI surfaces within the EHR, reduces context switching but demands rigorous testing.

IT teams must validate patient and encounter matching, user access controls, template mappings, mobile device workflows, and audit logging. They also need clear downtime procedures. If connectivity fails or an identity link breaks, a fallback workflow must exist. Consent management is another critical layer: patients must understand when ambient recording is active, and staff need clear guidance for cases where consent cannot be obtained, such as with a confused or critically ill patient.

Security is a shared responsibility. Microsoft provides encryption, data segregation for protected health information, and enterprise identity features, but the healthcare organization remains accountable for device management, staff training, and incident response. Retention policies for audio and generated text must be defined, and audit logs should be regularly reviewed.

For Patients

The potential upside is more attentive care. When a clinician isn’t typing, they can observe tone, body language, and subtle cues that might otherwise be missed. That could improve diagnostic accuracy and patient satisfaction, especially for older adults, people with communication challenges, or those in distress. However, ambient recording changes the nature of a medical visit. Trust depends on clear, operationalized consent—not just a checkbox buried in a form. Patients who decline must receive identical care without pressure.

How We Got to AI-Assisted ER Notes

Microsoft’s path to Dragon Copilot started with the 2021 acquisition of Nuance Communications, a leader in clinical speech recognition. Nuance’s Dragon Medical had been a staple for dictation for years, but the company had also invested in ambient clinical intelligence through its DAX product. After the acquisition, Microsoft folded those capabilities into its Copilot brand, layering on generative AI from its Azure OpenAI Service. The result is a tool that not only transcribes but also structures and summarizes clinical dialogues.

The backdrop is a well-documented crisis. Studies routinely show that physicians spend two hours on EHR and desk work for every hour of direct patient care—and often add “pajama time” at home to finish notes. Emergency departments, with their high volume, constant interruptions, and rapid patient turnover, amplify that burden. The COVID-19 pandemic accelerated burnout, leading many health systems to explore AI as a way to reclaim clinical focus.

Brown Health, under Chief Digital Information Officer Dr. Adam Landman, has positioned AI as a central lever for transformation. The system faces the same financial pressures as peers—rising costs, workforce shortages, and growing demand—and sees tools like Dragon Copilot as a way to improve both efficiency and care experience without simply asking clinicians to document more.

What Health Systems Should Do Now

For organizations evaluating ambient AI, Brown Health’s experience offers a playbook:

  • Start with a concrete use case. Emergency medicine and primary care, where documentation volume is highest, are natural entry points.
  • Run a controlled pilot. Test with willing providers, measure documentation time, note turnaround, clinician satisfaction, and—critically—error rates. Don’t rely on time saved alone; track whether edits are substantive or cosmetic.
  • Integrate early. Standalone tools can work for pilots, but plan for EHR embedding from the outset. This avoids later rework and ensures consistent workflows.
  • Build consent into clinical workflows. Train staff on how to explain ambient AI to patients and what to do when consent is not possible. Have a plan for pausing recording.
  • Monitor for bias. Evaluate speech recognition performance across accents, dialects, and clinical complexity. Check that note quality doesn’t differ by patient demographics.
  • Prepare for AI agents. If Brown Health’s broader agent strategy works, tools that summarize records, draft patient communications, and route tasks will follow. Establish governance now: define agent roles, permission boundaries, and escalation paths.

What’s Next

Ambient clinical documentation is poised to become standard. Epic and other EHR vendors are developing similar AI-powered assistance, and regulators are beginning to pay attention to how these tools are validated and disclosed. The next frontier is AI agents that not only document but also act—booking follow-ups, pre-filling orders, flagging care gaps. For IT leaders, the challenge will be to integrate these capabilities without creating a new layer of software complexity for already stretched clinicians. Success will belong to those who treat AI not as a productivity gadget but as infrastructure for a more humane healthcare system.