Great list. One underrated addition: model routing APIs (like LiteLLM or Portkey) deserve a spot because they're increasingly core infrastructure in multi-model production apps, not just a nice-to-have. In the projects we build, we treat the router layer as its own architectural decision — it's what lets you swap underlying models without breaking application logic, which matters a lot as the model landscape keeps shifting. Also worth calling out OpenAI's Assistants API as a double-edged sword: fast to prototype, but the vendor lock-in at scale is real and worth planning around early.