Agent Observability

A 6,500-run benchmark puts the framework cost spread at $0.47 per 1,000 runs and 98% of wall-clock on the model, while LangGraph deployments lose agent state 41.5% of the time a container is replaced. The losses sit at three other boundaries.