Performance of the tools doesn't really matter when you have a full LLM inference loop in between each tool call.

I still find it revolting they're writing this stuff in typescript.