prompt caching yay
claude has instilled new limits and a usage tracker, and today i learned the obvious thing that they are doing prompt caching. the first message i sent to opus cost 14% of my usage (normal, it's in a project with 100k of context) but my immediate follow up question only took 2% more.
which means i can't take my sweet time responding to these bots, gotta hustle like i do to respond to emails promptly...