Working on a shadow of myself
Lately I’ve been working on two small, private things around the same idea: an AI version of me that is more honest with me than I am.
The first is Djole’s Shadow, a chat that answers as me, but critically. Every reply has the same shape: what it notices, what I’m avoiding, and one next step. It also keeps notes about me. Anything I correct or write in myself is locked, and it can’t change it. Its own guesses start as low-confidence hypotheses, and when the memory is full, the oldest unlocked guess goes first.
I started it this week. It runs only on my machine, and most of it hasn’t been tried in a real conversation yet.
The second is an experiment with my mentor. There’s no chat. An agent looks at a few counts from my work, like active repos and closed tasks, plus one sentence I write each day. When one of the rules my mentor writes fires, it asks me one question, and gives no advice. We haven’t started the two weeks yet.
I don’t know yet if either of them helps. I’ll write again when I do.