Caching, batching, model routing, prompt compression — practical money-saving skills that cut AI API costs by 50-80%.