For the past year I’ve been quietly building a personal AI platform called OpenClaw — a stack of autonomous agents, cron jobs, local models, and channels into Telegram and Discord, all running natively on a Mac Mini in my flat. No cloud bills. No subscriptions. Just an M-series Apple Silicon box, 64 gigs of RAM, and a lot of evenings.

It started as curiosity. It turned into something I rely on every day. Roger, my main agent, sends me a morning brief at 7am, an evening recap at 6pm, watches my stock and crypto positions at 9pm, and runs about forty other cron jobs in the background. Hermes, his sibling, writes code with me through the day.

I’ve learned a lot building it. Most of what I learned came from getting it wrong first. So I’m writing it down.

What this site is

Five things I care about, roughly:

1. AI agents — what they actually are in 2026, where they break, and how to build them so they don’t.
2. OpenClaw — the architecture, the trade-offs, the cron jobs, the failures.
3. Claude Code — my daily driver. Hooks, sub-agents, MCP servers, workflows that stuck.
4. Codex — OpenAI’s CLI, free via ChatGPT OAuth. Underrated.
5. The future of working — what a job looks like when every engineer has ten agents.

What this site isn’t

A newsletter funnel. A course. A pitch deck for an AI startup I’m about to launch. I’m an IT engineer who likes building things. This is just where I write them down.

If any of that sounds interesting, the blog is the index. The about page has the longer story. Find me on GitHub if you want to see the code.

More soon.