Cost, Latency & the Quality Triangle

Every LLM decision trades three currencies at once: quality, cost, latency. A bigger model buys quality at more cost and more latency. Caching buys latency at staleness risk. A cheaper model buys budget at quality risk.

The design error is treating any one as free: 'just use the best model' spends the other two currencies without asking. Every production feature sits somewhere on the triangle, and 'where' is a product decision disguised as an engineering one.