Microsoft has launched a new solution designed to revolutionize how clinical AI tools are developed and deployed. The company’s healthcare The post Microsoft Opens AI “Store” for Healthcare Developers appeared first on The New Stack.
Microsoft has launched a new solution designed to revolutionize how clinical AI tools are developed and deployed.
The company’s healthcare agent orchestrator, announced today at Microsoft’s Build conference, represents a significant advancement in bringing AI capabilities into healthcare workflows while addressing the unique challenges of medical environments.
A Store, Not Just a Tool
Describing the orchestrator, Dr. Nigam Shah, professor of medicine and chief data scientist at Stanford Health Care, told The New Stack, “It’s not one tool. It’s a suite of agents that are available in the healthcare and life sciences catalog.”
Dr. Shah describes it as more like a store where developers can browse different capabilities: “I can go and say, ‘Where are your imaging agents, and where are your literature summarizer agents, and where are your clinical trial agents?’ So, I have a menu to pick from.”
This approach addresses one of healthcare’s fundamental challenges: the diversity of requirements across specialties. As Dr. Shah noted, “The things that a dermatologist needs in an outpatient setting versus the things that an ophthalmic surgeon needs in the operating room are dramatically different.”
Shah serves as tenured faculty in medicine and chief data scientist for Stanford’s healthcare system. He was originally trained as an MD and got involved in data science after the Human Genome Project sparked his interest. He teaches data science in medicine courses in both R and Python.
Shah had to learn Python and R, “and Perl and Lisp and a whole bunch of other programming languages,” he said. “During my PhD, my wife’s advisor took me under his wing and designed a custom curriculum for me to complete discrete math, data structures and algorithms, database systems and so on.”
Stanford Healthcare IT
Shah described the Stanford healthcare IT environment as consisting of “5,000 small businesses” with 50 clinical departments across 100 physical locations. It operates with approximately 1,300 different IT systems running on Microsoft Azure and has 123 different medical specialties, each with unique IT and infrastructure needs, he said.
Stanford uses Microsoft Azure as its primary infrastructure for all IT systems. The university just gained access to the Microsoft agent orchestrator platform two months ago.
From Months to Minutes
The immediate benefit for healthcare developers is dramatically accelerated development times. Tasks that previously required extensive custom coding can now be implemented by selecting pre-built agents from the catalog.
Timothy Keyes, an MD/PhD candidate working at Stanford’s healthcare system, found he could deploy agents in minutes rather than months, Dr Shah said. “Each and every one of those jobs would have been a three to six months development adventure for us,” said Dr. Shah, “but now I go to this catalog, the store, I can just check out an agent, and it’ll summarize the literature for me, and we can deploy it in less than 30 minutes.”
The Bridge Between Medicine and Computer Science
Meanwhile, Dr. Matthew Lungren, chief scientific officer for Microsoft Health and Life Sciences and former Stanford faculty member, identified a critical “translation gap” between domain knowledge in medicine and computer science that the orchestrator helps bridge.
“There’s a lot of domain knowledge in medicine, and then there’s a ton of domain knowledge in computer science. And just even speaking from the fundamentals on either side is still hard to do,” Lungren told The New Stack.
His journey into this field began almost accidentally in 2013 when he attended a seminar on ImageNet by Fei-Fei Li and Andrew Ng. “We look at a lot of images in healthcare,” he said he realized, setting him on a path to build solutions that could apply AI advances to medical practice.
Tumor Boards: A Perfect Use Case
Both Dr. Shah and Dr. Lungren highlight tumor boards — multidisciplinary meetings where specialists review complex cancer cases — as an ideal application for this technology. Overall, tumor boards typically involve radiologists, pathologists, surgeons, oncologists, genetic counselors and other specialists doing sophisticated analysis of vast patient data and knowledge to align on personalized care plans.
“Every year, 20 million people globally are diagnosed with cancer,” with hundreds of distinct tumor subtypes requiring specialized treatment protocols. Yet “less than 1% of these patients have access to personalized treatment plans” through tumor boards because of the immense preparation required.
Dr. Shah shared his personal experience from 2021 when he was diagnosed with a small tongue lesion. His treatment decision involved manual research by his surgeon and lab team to find similar cases — work that could be automated and expanded using the agent orchestrator.
“Clinicians spend between 1.5 to 2.5 hours per patient, meticulously reviewing imaging, pathology slides, clinical notes, and genomic data,” according to an American Society of Clinical Oncology (ASCO) study cited by Microsoft. The healthcare agent orchestrator aims to reduce this burden.
Key Capabilities
The orchestrator provides several key capabilities:
Multimodal data handling: Analyzes diverse healthcare data from imaging (Digital Imaging and Communications in Medicine files) and pathology to genomics and clinical notes. Seamless integration: Works within familiar Microsoft tools like Teams, Word and PowerPoint. Customization options: Developers can fine-tune each agent with their own models and data sources. Explainability features: Provides “trust but verify” mechanisms essential in healthcare. HIPAA compliance: Operates within secure Azure environments.
Dr. Lungren said the orchestrator delivers specialized agents for specific tasks: “The patient history agent leverages Universal Medical Abstraction to organize patient data chronologically. Manual work that can take experts over three hours happens in minutes.”
Early Adoption
Several leading institutions are already exploring the platform, including Stanford University, Johns Hopkins, Providence Genomics, Mass General Brigham and the University of Wisconsin School of Medicine.
In a statement, Dr. Mike Pfeffer, CIO of Stanford Healthcare, explained that “Stanford Medicine sees 4,000 tumor board patients a year, and our clinicians are already using foundation model generated summaries in tumor board meetings today. The new healthcare agent orchestrator has the power to streamline this existing workflow by reducing fragmentation.”
The Developer Experience
For developers, the orchestrator addresses key healthcare IT challenges. As Dr. Lungren noted, “Security and privacy are paramount concerns” in healthcare.
The solution operates within Azure tenants, keeping sensitive data behind secure “fences.” This approach eliminates what Dr. Shah calls “having to send stuff over the internet and then rely on unspecified protocols.”
The development experience is streamlined through integration with existing workflows. As Lungren said, “Most health systems leverage Windows-based enterprise technology at some level, even if it’s just for workplace productivity.”
Dr. Shah noted that Keyes, who has completed his PhD and is halfway through his MD, is working on his team as a software developer and data scientist, and Keyes was “through the roof” when he saw the orchestrator.
Keyes’ enthusiasm was because, “We on our campus are building a tool we call ChatEHR. You can imagine it does what it sounds like it does. It’s built using secure GPT running on our Azure tenant, and we had all these plans of, instead of just chatting with the record, ‘Could I recommend literature-derived evidence for this patient? Could I deliver real-world evidence derived from similar patients? Could I deliver an ability to search an image?’
“Now, each and every one of those jobs would have been a three to six months development adventure for us, but now I go to this catalog, the store, I can just check out an agent, and it’ll summarize the literature for me, and we can deploy it in less than 30 minutes.”
Microsoft said the solution’s orchestrating agentic capabilities can reason over complex EHR data and augment time-consuming tasks like building a chronological patient timeline, determining cancer stage leveraging specific reference guidelines, reviewing radiology and pathology images, synthesizing current medical literature, referencing treatment guidelines, surfacing relevant clinical trials and generating customized reports.
In addition, it enables developers to create, customize and fine-tune each agent with their own models, tools, instructions and data sources; test performance in a guided playground experience; and extend agents by setting up using Copilot Studio to integrate a Model Context Protocol (MCP) server with the sample implementation provided.
A Path To Democratizing AI in Healthcare
Perhaps most significantly, the healthcare agent orchestrator represents a move to help democratize healthcare AI development, the doctors said.
Dr. Shah compares it to how cell phones evolved: “When cell phones got invented, the calls used to cost eight bucks a minute, and very few people could afford it. Now, everybody has a cell phone, but that happened because that capability is now performed at a much lower price point because of automation and software.”
Similarly, the orchestrator could expand access to advanced AI capabilities across healthcare systems of varying sizes and resources, potentially addressing a critical gap in the healthcare workforce.
Dr. Lungren said he believes this technology “will save time before it saves lives.” By reducing “undifferentiated heavy lifting” in healthcare workflows, it promises to free clinical staff to focus on direct patient care.
Looking Forward
The healthcare agent orchestrator is available now via the Azure AI Foundry Agent Catalog. It features pre-configured agents with multiagent orchestration and open source customization options that allow developers and researchers to build agents that coordinate multidisciplinary, multimodal healthcare data workflows (such as tumor boards) and streamline deployment into healthcare enterprise productivity tools (such as Microsoft Teams and Word).
While currently intended for research and development rather than direct clinical application, the orchestrator’s potential impact on healthcare delivery is substantial.
Indeed, Microsoft said the healthcare agent orchestrator is intended for research and development use. It is not designed or intended to be deployed in clinical settings as is, nor is it intended for use in the diagnosis or treatment of any health or medical condition, and its performance for such purposes has not been established.
However, the platform represents a significant step toward integrating AI capabilities into healthcare workflows in ways that respect the complexity and diversity of medical practice, Dr. Lungren said.