How do teams prevent duplicate LLM API calls and token waste?

I'm curious how teams running LLM-heavy applications handle duplicate or redundant API calls in production.

While experimenting with LLM APIs, I noticed that the same prompt can sometimes be sent repeatedly across different parts of an application, which leads to unnecessary token usage and higher API costs.

For teams using OpenAI, Anthropic, or similar APIs in production: How do you currently detect or prevent duplicate prompts or redundant calls? Do you rely on logging and dashboards, caching layers, internal proxy services, or something else? Or is this generally considered a minor issue that most teams just accept as part of normal usage?