I built a similar autonomous loop using LangGraph for a publishing backend and the raw API costs were significantly higher than $200. The subscription model likely has opaque usage limits that trigger fairly quickly under that kind of load. For a bootstrapped setup I usually find the predictability of the API bill worth the premium over hitting a black box limit.