Agentic AI, in production, without getting burned

My job at Accenture LearnVantage is changing how a very large organization works with AI: workforce transformation across adoption, enablement and the governance that has to arrive with it. The engineering is the easy half. Thirty years from CTO to individual contributor taught me that a control nobody follows is not a control, which is why Agentic Covenants is written as a matrix of what each layer can actually enforce rather than as a policy document. The MCP Dev Summit Toronto keynote is the working example: about 750,000 people, one protocol, and the question of what governs it. The writing here covers the conversations that come before a rollout, starting with the four-hour executive AI workshop problem.