Microsoft is betting that Dragon Copilot, its ambient clinical documentation tool, becomes the operating system for clinical decision-making. The pitch: instead of an AI that writes down what happened after you've seen a patient, Dragon Copilot surfaces evidence-based recommendations mid-encounter. More than 100,000 clinicians already use it. Microsoft expects hundreds of plug-ins from third-party developers within a year.
But the deployment data tells a different story. Forty-three percent of healthcare organizations are piloting or testing agentic AI, according to research Microsoft published in February with The Health Management Academy. Only 3 percent have actually deployed agents in live clinical workflows. Sixty percent of healthcare executives believe agentic AI will meaningfully improve or disrupt the provider-patient experience. The belief-to-action gap is enormous and it's the gap Microsoft is trying to close with Dragon Copilot's marketplace model.
The shift matters because ambient documentation was always the entry point, not the destination. Dragon Copilot's origin story is speech-to-text in exam rooms: doctors talk, the AI types. At HIMSS26 in March, Microsoft showed how far the ambition has stretched. Dragon Copilot now integrates with Microsoft 365 Copilot through a layer called Work IQ, synthesizing clinical data from inside the EHR with operational data from emails, Teams chats, and SharePoint. A clinician can query a patient's lab results, cross-reference a hospital policy document, and check their own schedule in a single natural-language exchange.
"This is what historically we've done with DMO — with dictation at the cursor, where you talk and then we insert text into wherever your cursor is," Kenn Harper, general manager of Dragon Product at Microsoft Health and Life Sciences, told Healthcare IT News at HIMSS26. "It's leveraging that same intelligence, but we're supercharging it with Copilot, where we now can allow you to basically speak a natural language."
The more consequential expansion is the plug-in ecosystem. Microsoft Marketplace lets health systems deploy third-party AI apps directly inside the Dragon Copilot workflow. At launch, partners include Canary Speech, which analyzes voice for anxiety and depression markers during appointments; Humata Health, which handles prior authorization by surfacing payer guidelines at the point of care; Optum; and Regard. Regard, which Sentara Health is integrating, surfaces comorbidities in real time by combining Dragon's ambient conversation capture with its own diagnostic inference engine.
"By combining Dragon's ambient conversation capture with Regard's ability to surface key insights from data, we expect to help our clinicians identify comorbidities and relevant diagnoses in real time without adding steps to their workflow," said Dr. Joseph Evans, vice president and chief health information officer at Sentara Health.
The nurse workflow is new territory. Mercy health system is running Dragon Copilot for med-surg nurses, who can document hands-free at the bedside. "I can say that without a doubt, using Dragon Copilot has significantly reduced the time that I'm focused and worrying about sitting down and getting my charting done behind the computer," said Christine Dupire, a Mercy nurse, in Microsoft's announcement. Radiology gets a preview integration with PowerScribe One, automating summary generation of prior reports so radiologists can focus on interpreting new images.
The structural explanation for the 3-percent deployment figure surfaces in Microsoft's agentic AI readiness research: qualitative interviews with senior healthcare executives revealed a consistent pattern — organizations view agentic AI as a strategic end state, but believe that end state depends heavily on progress in workforce readiness, governance frameworks, and data infrastructure.
The broader platform question is whether Microsoft's marketplace strategy succeeds where earlier healthcare AI bundling efforts failed. The pitch is coherent: turn Dragon Copilot into a distribution channel for clinical AI, the way iOS became a distribution channel for mobile apps. The installed base of 100,000 clinicians is real. The plug-in model reduces friction for health systems that want specific capabilities without negotiating separate vendor contracts. The counterargument: healthcare AI adoption has historically followed the EHR vendor more than the productivity tool. Epic, Oracle Health, and athenahealth all have ambient documentation competing for the same workflows. But the 3-percent live deployment figure applies to the entire industry.
What to watch: whether the hundreds of expected Marketplace plug-ins actually materialize in the next 12 months, and whether early partners like Regard can demonstrate clinical outcome improvements that justify the workflow integration. Microsoft's ability to execute on the platform vision depends less on its own AI capabilities and more on whether the developer ecosystem treats Dragon Copilot as worth building on.
Dragon Copilot is currently available in the United States, Canada, the United Kingdom, Ireland, France, Germany, Austria, Belgium, and the Netherlands, and captures clinical conversations in 58 languages.