The happiest thing to do with AI is to fully utilize the weekly credit limit of all two subscribed $200 accounts in the minutes before resetting, and use it to create something that solves problems, rather than just a demo. I sent both Claude and Codex resets on Wednesday this time. Codex quickly ran out of GPT-6 Astra Medium and was mainly used to help me analyze requirements and review code, which is a more detailed task. In order to use it more, I did not turn on Fast mode. Claude Code uses my favorite Opus 5.5 high model, which is very, very durable. I truly feel like it has reached the title of Max. Now, Codex Pro has a Pro mini feel, unused, but fortunately it often comes with Reset. At the same time, I am very grateful to Codex because he gave me a 6-month open source subscription, which allows me to do some interesting things. The reason why I like Opus 5.5 is that it really helped me. Without it, the Mole Windows version wouldn't have formed so quickly and could have been tested by me. In order to test it, I even built a ThinkPad X1, which is a very lightweight and decent machine. However, my main developer still developed Windows CS native code on Mac and used Parallels Desktop, a very useful virtual machine, to apply it. I also purchased a one-year subscription and used Windows very smoothly on it. Moreover, my Agent can use many software inside to help me test the effect, which really surprised me. The price of over 300 yuan per year is very worthwhile. In addition, I have also used the $200 monthly account that was given to me by Cursor, which is very good. I had a credit limit of $10000 before, but it was used up in a month. I did a lot of code for the well-known Kami and Waza Mole Mac versions. Recently, I used his corresponding Grow Bot to help me with some high-quality global user growth, which is very interesting. Then, I used Grow 4.7 Extra High Fast to help me solve various open source issues and PR problems. After a month, I actually used it almost as much, very quickly. Although not as smart as the first two, I feel that I can rank third, which is also very good. After the acquisition of cursor by my favorite SpaceX, there is a feeling of 1+1>>2 on both sides, which is also very interesting. It's a bit rambling, but what I want to express is that it's really interesting to be able to use the AI model you subscribed to in practical scenarios without any waste before resetting. In addition, I also dislike some companies or teams' so-called token usage ranking, which focuses on who uses how many tokens instead of what they do with them. It's actually a case of putting the cart before the horse and easily guiding engineers to behave in a bad way. The use of AI is more interesting when it is fully utilized. Writing fewer demos and producing more products is more interesting.