This was a really useful breakdown. The metrics / logs / traces distinction also made the topic much easier to understand. I’m curious whether you think LLM observability will eventually become a default layer in AI products, the same way application monitoring is now.