Essays and notes.
Writing about building, deploying, and operating AI agent harnesses in businesses.
-
Neutron is open source
I built an agent harness from scratch, after pushing OpenClaw to its limits and not feeling satisfied. Ninety days later it runs my businesses and my life. Today it's open source: Apache 2.0, self-hosted, your data on your own disk, your own Claude subscription. Here's what's in the repo and how to stand it up.
-
What the agents actually did
Four months ago I started running my businesses and side projects exclusively through my AI agent harness, instead of doing the work myself. This is an account of what actually happened: what got automated, what got shipped, the wow moments, the challenges, and the parts that still need human judgment.
-
How I orchestrate twenty Claude Code sessions from my phone
This is a hands on guide to building an agent harness to orchestrate a fleet of Claude Code instances. We start from a single Claude Code session in a tmux pane and build up to a fleet of long-running sessions, one per project, reachable from a chat app through a routing gateway, with an MCP reply tool, long-term memory, reminders, and background workers. The channel layer is just an adapter; I wire it to Telegram because that's how I often run mine. Real code, real file paths, the actual architecture I run every day