Over the last two years, we've encouraged employees to teach AI how work gets done. With Copilot, Agents, and Skills, that idea is finally becoming reality. Teams are capturing expertise, packaging processes, and sharing knowledge in ways that were difficult just a few years ago.

What surprised us was not how quickly people started creating skills. It was what happened when those skills became successful.

A useful skill rarely stays where it started. It gets shared, adapted, copied, improved, and passed between teams. All of that happens for good reasons, but after a while a simple question becomes difficult to answer: Which version should people actually trust?

In this session, we'll discuss what we've learned from customer conversations, community discussions, and our own experiments with skills governance. We'll explore why the next challenge in enterprise AI isn't creating more skills, but distributing them in a way that preserves trust, consistency, and ownership as adoption grows.

We'll also introduce the idea of enterprise-managed and sealed skills: skills that can be used across the organization while remaining connected to a centrally governed source. Because once AI becomes part of everyday work, the conversation quickly shifts from capability to trust, and from creation to distribution.

If you're investing in Copilot, agents, or enterprise AI, this session is about what happens when adoption starts going well.

In September 26 we do not know, what will be in March, April 2027. We started to talk about skills governance in July and we need to keep this topic and session pdated as new capabilities will appear. Togehter with some developers of Microsoft we discussed our ideas about Skills governance and sealed skills.