How Do You Load-Test an Application That Depends on AI APIs?

When a third-party LLM API sits in your critical path, classic load testing breaks. Rate limits, per-token billing, and wild latency variance turn a traffic spike into an outage and a surprise invoice at the same time.