If you don't run your own LLM proxy, how do you know whats going on?

The article argues that running your own LLM proxy is necessary to understand what is happening with your AI workloads.