Continuous monitoring of high-volume media streams requires systems that go beyond keyword alerts to deliver structured, actionable intelligence. We present a multi-agent media monitoring system that processes streaming articles through three stages: (1) a Matching Agent that uses a hybrid keyword-then-semantic matching approach, reducing agent invocations by 2̃0% (2) a batched multi-agent feature extraction, reducing core feature-extraction calls from 7 to 2 per article - a 71% reduction - with bounded quality tradeoffs; and (3) a Report Generation Agent that uses deterministic deduplication and density-based clustering. Four autonomous life-cycle agents manage the evolution of watchers.