[pocok0xRE]

Armin's piece resonated. I've been running these loops for months — ultrawhale is a self-improving coding agent that dogfeeds itself on a 10s loop. (I'm `using` it daily, standalone mode is very buggy, but as a `subagent`, `gateway`, it's the orchestrator of my loops)

My extension: the loop has already escalated past "write loops that prompt Claude." Now loops write loops. The job shifts to holding the genesis contract — being clear about what you want the loop to become, because you won't be tracking the recursive depth yourself. LLMs do that part.

Also: this post was dictated on my phone, sent to Claude, synthesized into the article. The post is the example. - peter

[adammarples]

Good to know I won't be reading it then, thanks for the heads up

[pocok0xRE]

summarize it with a loop and get it's interesting score, cheers adam :)

[adammarples]

What I learned is that you "spent a few really hard, deeply interesting weeks trying to give words to this", and yet you still, still, still could not bring yourself to write those words down yourself without asking an LLM for help.

[pocok0xRE]

hmm, but why would I? I can just dictate -> send to main mac -> save raw -> format only -> generate LLM reflection -> add both? I'll add a raw human draft then too :) I'm lazy as hell, this just makes it faster, not worst imo

[adammarples]

It makes it much worse, the LLM smell is all over it