Always seeking continuous improvement and learning. Building, experimenting, and growing, one step at a time.
I dug into the Hermes agent's self-learning claims at the code level. What I found clarified what's possible, what's marketing, and what the real frontier looks like.
Group chats, DMs, and heartbeats all create separate sessions inside the same agent. When you don't know that, things break in confusing ways.
My OpenClaw agent launches Claude Code sessions, answers most of its questions autonomously, and only taps me when it needs my judgment. Here's how the whole system works — the SDK, the wake mechanism, and why I had to teach my agent to stop micromanaging.
A single commit to my agent's instruction files turned it from a process-document generator into a machine that ships real work. Here's what I changed and why it worked.
How I split a single overwhelmed AI agent into three focused agents — each running its own business — on one Raspberry Pi.
Why I built a dependency-aware issue tracker for my personal agents, and how it closed the loop between planning and shipped outcomes.
A practical workflow for turning notes into shipped outcomes across phone, laptop, and Raspberry Pi using Obsidian and OpenClaw.
How a Raspberry Pi, Tailscale, Claude Code, and OpenClaw turned into a private 24/7 execution engine for work and life admin.
How editing OpenClaw's default workspace markdown files changed agent behavior, improved proactivity, and cut token waste in real usage.
How access to high-tier AI models is creating a new economic and cognitive gradient.
A deep dive into how I investigated a system freeze, traced it to memory thrashing, and built a multi-layer defense system.
Why ambition alone often produces mediocre outcomes — practical steps to convert ambition into focused action and measurable progress.