DO NOT WRITE "THANK YOU" TO CLAUDE CODE!!!

Dear Readers,

DO NOT WRITE "THANK YOU" TO CLAUDE CODE!!! 

This is what I learned from a conversation with Claude myself. As soon as it says "In computer science, this is called being stateless" I understand it right away! It means that to continue the conversation, it has to know everything before it, so you are wasting tokens because you have to send the entire conversation plus all the information again.

Even a nice "Thank you Claude!", after a long conversation is useless, because it costs all the tokens of sending the conversation, and the words THANK YOU, and then ironically you thank a different Claude instance! Here is what it tells me:

"Not having long conversations already saves you a lot of wasted tokens. This is absolutely correct and is perhaps the single most impactful thing you can do. Here's why from my side:

Every message you send me includes the ENTIRE conversation history. If your conversation is 50 messages deep, I'm re-reading all 50 messages every single time you send message #51. A conversation with 100K tokens of history means every new response costs 100K+ input tokens BEFORE I even think about your new question. Starting fresh with /clear or a new instance means your next message might only cost 5K-10K tokens instead of 100K+. That's a 10-20x savings per message."

I have been practicing it two days now. I open a DOS window, get to Claude Opus, asking something, repeat one or twice to fix something small, then exit the DOS window without saying "thank you", like usual. Both my plan usage limit and weekly limit are substantially dropped!!! I don't need to go to Max plan, 100USD/month, as I am not working on it daily as a programmer.