yea, you might be notice it, my 4 previous blog is actually using AI, and yes I can’t deny it, I’m actually trying to understand “what if my journal is written by AI”, and you already knew, it turn out to be full with AI slop.
and thats what happen since the April until last July, I feel the worse of using AI, the AI fatigue is happening.
what I mean by AI fatigue, the condition when you feel overwhelming by the condition because it has kinda X Y Z optional execution, and people still kinda talk about X Y Z is a thing, and you kinda vomited by the overwhelming.
So as I’m writing this now, we still debating over which “harness” AI would be best.
and someone decided to start this war
https://x.com/sayodotfun/status/2102098349746971103
you see, since 2025 I still use claude code, and the best way to use AI harness is you built it up from scratch, like arch, or nix you control everything.
why do you think so ? by the basic of need AI harness could be handle all of the execution request from the user, here I will show you how usually AI Harness will work
so as you could see, usually the AI harness will do some acrobatic dumb from every layer thinking its the revolution, while you culd actually just doing it from the scratch
but like the wise man said
“Don’t be a dumb by doing everything from a scratch” - wise man
about a year already people still talk about the “System prompt” leak, kinda like when you pick one of the AI harness, you will see the reason why people love to inventing another same tools.
I believe they wanted to make a better AI harness by crafting the system prompt, making the execution layer more “beautifull” in the eye of the user, and well I believe it just some kinda another AI slop.
around this time what I did is using claude code + CCS (claude switch) + SKILLS.md + 8 custom MCP, ohhh I feel happy.
So the day I afraid is come, the mf AI slop from claude, the safeguard BS
look at this image, what happen is, I mind my own business, working with claude account, doing testing IDOR, bypass permissions into live production, because they asking me to do so, and sometime ssh-ing into server to understand the request,
and this mothef think I do some kinda like burning the whole building.
https://x.com/BukanYahya/status/2100168820602351924
I GOT THE SAFEGUARD BY THE CLAUDE.
ok ok calm down, I only need go to link, and then register as “the registered tester”
from my point of mind this is kinda like “micro-controlling” user whenever they wanted to do anything.
and you will be got flagged as “dangerous activity” when the AI slop think I’m doing the dangerous activity
so 14 days later, I feed up with this,
I throw out the account, put it into hermes jail only working with telegram.
and starting my journey to a “minimalist”, and this is the day I decided using π as my harness
and you should look at this “System Prompt” Leak from claude code https://github.com/asgeirtj/system_prompts_leaks/blob/main/Anthropic/official/2026-09-22-claude-opus-5.5.md
and now you compare it with pi
https://github.com/asgeirtj/system_prompts_leaks/blob/main/Pi/instructions.md
and that is why I move to pi
https://x.com/BukanYahya/status/2104180441356104093
and makes my day more peacefull without any stupid MCP, just a peacefull mind
conclusion ?
you will never notice the AI fatigue happen, you will only thinking this and that will never enough until you kinda tired moves from part to another part pick something that ease your mind
for example, I move from visual studio that has garbage visual into VIM