Agentic AI, in production, without getting burned

I let an agent build an entire internal developer platform and wrote up what that actually looks like. I built KubeQuest and a D&D virtual tabletop with agents doing much of the typing, and I keep a Lancer campaign running on the side. Every episode of The AI Inevitable takes one falsifiable claim about AI and checks it against the papers. The serious version of all this is Agentic Covenants: if the agent decides to misbehave, what actually stops it? The playful version is most of what is on this page. Thirty years of infrastructure is what keeps the two honest.