BLOG
Notes on e-commerce, front-end, leadership, and coding with AI. Straight to the point.
════════════════════════════════════════════════════
I moved all my agent work onto my own server
Every Claude Code session I start runs on the mini PC in my house, reached from my phone, my notebook or my desktop. It is the same machine running my 24 containers, so the agent is already inside the thing it is changing.
48 hours rebuilding my home server
Plex said it could not connect securely. The cause was an SSD refusing writes while SMART reported PASSED, and the audit at the end found a backup that had never run once.
Why I switched to Orca
I tried Conductor and Superset before landing on Orca, an open source ADE for running coding agents. What pulled me in was the remote setup: an agent on a box over SSH, a real terminal that keeps running when the connection drops.
I stopped picking AI models by hand
Every Saturday my agent researches which models people are running, prices them, and compares that against my config. I read the report and decide.
Why I dropped paid dictation apps for FreeFlow
Superwhisper and Wispr Flow replaced by FreeFlow, an open-source project that calls a Whisper API instead of running a local model.
TakeLab: shipping software outside my stack
The spec I wrote by hand, the planning session before any code, and why the extension watches Bitwig's transport instead of the keyboard.