Blog
I Ran Qwen3.8 at 262K Context on a 96 GB Mac Studio28 Sept 2026How DwarfStar and mac-studio-server made a 262K-token Qwen3.8 coding session practical on a 96 GB M3 Ultra. I Built an MCP Server for My LG TV1 Sept 2026A local MCP server with 51 typed TV tools, exact YouTube playback, and a hidden screenshot API. How to Interrupt an AI Agent28 Aug 2026Your agent is busy and you just sent a correction. What a turn is, how Claude and Codex each let you steer one that is already running, and why asking whether a turn is running is the wrong question. I Wanted My Wife to Use a Finance Bot. I Didn't Write Any of It.26 Apr 2026From a 9pm voice memo to a family member using the bot at breakfast — what actually happened overnight, and what I was doing while it happened. One Workflow, Three Jobs: How We Built a Reusable AI Review System16 Apr 2026How one 146-line workflow handles design review, code review, and full pipeline review — with real bugs caught in production. Phleet Architecture Deep Dive11 Apr 2026How I built a multi-agent system that runs on a Mac Studio — containers, RabbitMQ, Temporal workflows, and a JSON-driven workflow engine.