For my third year speaking at M365 Community Days NYC on enterprise AI development, I’m bringing the third act. Ownership.

Enterprise teams now need to answer harder questions. Where does the agent run? What can it access? Which skills define its behavior? What data boundaries does it respect? What test proves the behavior still works after the next change?

In this live technical session, live, we will inspect real skill files, run validation checks, execute behavioral tests, and write a new failure-mode eval for a privacy boundary. The architecture uses OpenClaw as the runtime, Tula skills for inspectable domain behavior, Microsoft infrastructure for the enterprise integration layer, and Waza evals for the verification harness.

The goal is simple. Move from impressive Agent demos to owned, inspectable, and verifiable agents that enterprise teams can actually operate.