I'm currently in the process of taking a lot of my data logbooks, chat exports, and reasoning traces and creating my own brain, which I'm then gonna root AI through. Essentially my own completely custom bespoke brain, which works in the same way a human brain does in regards to memory and promotion of thoughts and storage, etc. Been doing some early work on this and seen some good outputs so far. Just wanted to come on the community to see whether anyone has done similar and would kindly offer advice/be okay with me throwing questions at them in Dms ;]

Went back through an old folder of mine this week. It carries the full scaffold, research through drafts through audits, brief, spec, build, runner, and a routing table in CLAUDE.md that sends each kind of request to its stage. The thing it actually produces is one script doing one deterministic job. Every stage except the build is empty. The wasted directories don't bother me. What bothers me is that the routing table still promises the model something is in there. When a request matches an empty stage, the model doesn't come back with "nothing here". It fills the space with something plausible and keeps going, and the only reason I caught it is that the answer read a little too smooth for a folder with nothing in it. So, two things I'd like a read on from people further along than me. Do you keep the full scaffold on single-purpose folders for consistency across a workspace, or strip it back to only what actually holds files? And if you keep it, what makes an empty stage announce itself instead of being quietly routed into?

I keep running into issues with organizing my file structure. I keep having to tell my chats to review the ICM templates and examples to make sure workflows are structured accordingly, but I still feel like i'm all over the place. Has anyone else had similar issues or suggestions for assistance. It feels like it takes hours just to create one workflow, let alone test it🫠. I already have a workflow and skills for creating the files and folder organization but still having issue!

The more I spend time (tokens) prompting, the more it makes me think about how we all start getting prompted from the womb, every day, at all time. Some prompts make it to the super long term memory (sub conscious, trauma, conditioning)... some prompts are abvious, others are abstracted. The parallel between psychology and AI becomes more and more evident the more I work with my agents. It also makes me wonder who is prompting who, I my self start thinking more like an AI, or is that how developers think? It's definitely shaping me as much as I am shaping my agents. Not sure where this post is taking me, but I wanted to share these thoughts that emerge in my morning walks. Not many people around me IRL would understand what I am trying to convey... do you get it? would love to hear your thoughts.

So I don’t have much evidence on the quality of the output of Opus 5.5 yet. It does seem like it’s fixed the disaster of yappy and overthinking Opus 5. The one thing that I have noticed is that it’s light years faster than any previous Opus model. Anyone else seeing that too?