Instead of handing nurses another AI tool built by outside engineers, Mayo Clinic decided to ask its own nursing staff what they would build if given the chance. The result was an internal challenge that drew more than 450 submissions from nurses proposing largely autonomous AI agents to help with their own workflows, with 33 ideas advancing to a second round where finalists are now working directly with Microsoft engineers to build and refine prototypes.
A Challenge Built From the Bedside Up
The premise of Mayo’s challenge inverted the usual order of hospital technology rollouts, where IT and informatics departments typically select software and then train frontline staff to use it. Here, nurses themselves identified the problems worth solving and pitched specific agent concepts, with the health system relying on frontline knowledge of daily friction points rather than top-down assumptions about where AI could help most. One of the ideas that advanced to the finalist round is a tool that helps nurses quickly locate internal policies and procedures on the fly, a seemingly small but persistent pain point in large hospital systems where protocols are scattered across multiple reference systems.
How the Ideas Are Judged
Mayo built a formal review process around the submissions. Judges, including nursing executives, governance leaders and front-line staff, evaluated proposals on workflow and department alignment, risk, and implementation readiness, a structure meant to filter out ideas that sound appealing in the abstract but would be difficult or unsafe to deploy at scale across a large academic health system. After the finalist round, the ultimate decision will not rest solely with executives or the judging panel: the entire nursing department is set to vote on the final winners, giving rank-and-file staff direct influence over which agents actually get built out and deployed.
Building on an Existing Foundation
The nurse-led challenge did not emerge in a vacuum. Mayo’s nursing department had already built the Nurse Virtual Assistant, a generative AI tool developed entirely in-house and tested in a research study approved by Mayo’s institutional review board before being expanded to more than 9,600 nurses across inpatient and emergency department units. That tool surfaces a nurse-specific patient summary inside a single tab of the electronic health record, with direct links to clinical references nurses consult constantly, such as Lippincott procedure guides and IV administration protocols. Mayo has also been deepening its relationship with Microsoft more broadly; the two organizations announced in June 2026 that they would collaborate on a frontier AI model for healthcare, a separate, larger effort that sits alongside the nurse-driven agent challenge rather than replacing it.
Two Views on Letting Clinicians Design Their Own Tools
Supporters of Mayo’s approach argue that bottom-up design solves a problem that has dogged health IT for years: tools built without deep input from the people who use them every day often go unused or get worked around once deployed. By sourcing ideas directly from nurses and then pairing finalists with Microsoft’s engineering resources, Mayo is betting that frontline ownership will produce agents nurses actually trust and adopt. But governance and informatics specialists caution that scaling from 450 raw ideas to a handful of production-ready autonomous agents is a far harder engineering and safety challenge than it sounds, particularly because these are described as largely autonomous tools rather than simple reference lookups. Agentic AI systems that can take actions or make recommendations with limited human review carry higher stakes for error than passive information tools, and skeptics note that enthusiasm from a popular internal vote does not substitute for rigorous clinical validation, especially if multiple winning agents are deployed simultaneously across different units with different workflows.
What Happens After the Vote
Mayo has not disclosed a timeline for when the nursing department vote will conclude or when winning agents might reach production use, nor has it detailed what ongoing monitoring or risk review will apply once tools move from prototype to the floor. The approach nonetheless represents a notable experiment in how large health systems might crowdsource AI innovation from clinical staff rather than relying solely on vendor pitches or centralized IT roadmaps. If even a handful of the 33 finalist ideas prove durable in daily use, Mayo’s challenge could become a template other academic medical centers look to copy as they search for ways to make AI adoption feel less like something done to nurses and more like something built by them.
Photo: Tima Miroshnichenko / PEXELS via Pexels
