Stop duct-taping AI: Building a system of context for production

Right now, engineering teams are duct-taping AI into production with fragile scripts and framework spaghetti that no platform team can safely operate. The challenge is not generating answers; it is governing, observing, and constraining AI agents in real infrastructure. This talk introduces the "System of Context," a dedicated runtime layer that connects models, MCP tools, telemetry, and workflows. The session walks through how this pattern works using a real multi-agent Kubernetes SRE orchestrator, composed entirely from declarative config, with Prometheus queries, incident triage, and full OpenTelemetry tracing. Attendees will see how platform teams can ship AI agents they can actually govern: bounded, inspectable, and observable from day one.