What the keynote covered
John Siefert, Dewain Robinson and Giuseppe Ianni opened the program. Siefert asked attendees to "double down on the critical thinking aspect" of AI. Later, Jennifer Harris and Shane Young of TMC Global argued for curiosity over fear. Harris summed up the governance balance this way: "Get your right governance, but don't wait."
The rest of the keynote was customer and partner stories. These are speaker claims reported by Cloud Wars. I found no independent corroboration of the session details.
Vibe coding meets the clinical floor
Olof Hedin of Lovable said nurses at HCA Healthcare, a 189-hospital Tennessee system, are vibe coding their own interfaces. He said these connect to the patient record system and to Palantir. Hedin also cited power-wheelchair maker Permobil, whose CEO reportedly uses Lovable.
The report does not say what data-access controls or production safeguards sit behind the nurse-built interfaces. Treat the claim as a speaker's illustration, not a verified deployment.
Why Copilot Managed Runtime is the real story
Microsoft's own announcement fills in the context. Per its September 25 post, Copilot Managed Runtime entered public preview as a platform that runs code inside the Microsoft 365 tenant boundary under IT governance. Third-party tools can use an SDK. Lovable packages the app with Microsoft's Copilot Managed Runtime SDK and deploys it into your Microsoft Entra tenant. IT manages it like any other Microsoft app, and it shows up in the company's app inventory.
Independent coverage describes the mechanics:
- Apps inherit Entra identity, Microsoft 365 sharing controls and organizational policies for connectors, data access and endpoints.
- They get Git-backed source control, and a new Apps section in the Microsoft 365 admin center inventories every app published to the runtime, with usage and health visible and the ability to switch apps off.
- Teams can preview and refine a new version of an app while the current version stays live for users.
- Lovable says Microsoft 365 connectors, Fabric and Microsoft sign-in for built apps are available on every Lovable plan, while workspace sign-in with Entra ID is included on Business and Enterprise.
What admins should check
Microsoft's Learn documentation, last updated September 29, labels the governance settings as preview and subject to change. Based on that documentation:
- Each maker is routed into a personal developer environment. A default environment group applies four rules: routing, connectors and MCP servers, sharing, and content security policy.
- The default connector allow list is limited to Microsoft first-party connectors that use Entra ID authentication. Third-party and custom connectors are excluded by default. Open-ended actions, such as sending arbitrary HTTP requests, are blocked even on approved connectors.
- External artifact deployment is off by default. Microsoft states that the organization is responsible for validating externally built artifacts and securing its build pipeline.
- Switching the connector policy from Microsoft-curated to full control stops automatic updates from Microsoft. Existing apps might be affected by later changes.
- The routing rule for Copilot Managed Runtime cannot be turned off or deleted.
The preview label is the caveat. Governance does not by itself make a nurse-built clinical interface compliant. It gives IT a place to see, limit and switch off apps, but data sensitivity and validation remain the organization's job.
Dynamics transformation lessons
Two Microsoft Dynamics customer stories carried the change-management theme:
- Everon: The fire, safety and security firm was spun out of ADT. It was given 12 months to replace every system without interrupting service. It chose Dynamics systems including ERP and CRM. CIO Ryan Fritts said no one resigned during the 12 months, even though some employees worked 100-hour weeks. RSM's Mike Gillis credited a shared vision longer than the project itself.
- Barry-Wehmiller: Bonnie Duddleston described a Dynamics 365 and Truvio transformation at the 12,000-employee manufacturer. She said transformation depends on people understanding the value of the tools and choosing to use them. She also said clean, consistent data lays the foundation for future AI value.
The report gives no measured outcomes or application-level detail for either project. The lessons are about people and process, not technical benchmarks.
Microsoft's Georg Glantschnig, CVP for Dynamics 365 Agentic ERP Applications, offered a three-part framework:
- Rethink whole cross-functional workflows, not individual tasks.
- Put governance in place so AI works within a system of trust.
- Choose a process, define what success looks like, then learn and scale.
Agents as temporary teammates
Dona Sarkar and Leon Welicki, who leads Power Platform at Microsoft, described a Teams conversation with two humans and two agents that helped prepare their session. They also demoed an AI coding agent improving accessibility in an existing Power App. Welicki said organizations will be able to bring agents into apps for tasks like coding, security reviews and performance analysis, then deactivate them. His line was "Agents in when you need them, out when you don't."
The word "will" matters there. The report does not name the agent, give its availability or share evaluation results.
Microsoft's Power Apps documentation on generative pages is a useful check. It describes an accessibility assistant that scans generated code and offers an auto-fix. It also says the maker is ultimately responsible for validating the code against organizational standards and compliance requirements. That documentation covers model-driven apps, not necessarily the keynote demo. It does show that an AI accessibility pass is a first screen, not certification.
Practical takeaways
- Pilot with intent. Pick one workflow and define success before you pick a tool.
- Review your preview settings. Check the Apps experience in the Microsoft 365 admin center and the environment group rules before makers start publishing.
- Keep the external-artifact default off unless you have a pipeline you trust.
- Budget for people. Both Dynamics stories rest on employee buy-in.
- Fix your data first. AI on top of inconsistent data tends to produce confident inconsistency.
- Test AI-assisted output for accessibility and security with humans in the loop.
Bottom line
The keynote's tone was optimistic, and its examples were largely vendor and partner testimony. The substantive news for IT is the Lovable and Copilot Managed Runtime pairing. It is a preview that gives administrators a more consistent control point for apps people were going to build anyway. Whether it holds up in regulated settings like hospitals will depend on tenant configuration and testing, not on stage anecdotes.