Coding

I've run over 11,000 AI agent tasks this year with a 99.4% success rate. Here's what actually works, what fails spectacularly, and why boring agents beat brilliant ones.

Eighty lines of TypeScript and an MCP server turn Terminal Bridge into a full AI terminal integration. Here's the code, the patterns, and the hard lesson about terminal naming I learned at 2 AM.

Everything you need to know to play Flip Match (Memory Snap) — on Alexa or the web. Chess-style coordinates, three difficulties, four themes, and streak scoring.

Everything you need to know to play Zoo Simulator — on Alexa or the web. Habitats, animals, breeding, staff, the 7-phase daily simulation, and Golden Tickets.

Everything you need to know to play Blood Lineage — on Alexa or the web. Kingdoms, military, diplomacy, dynasty succession, and the three game modes.

Everything you need to know to play Murder Mystery — on Alexa or the web. Investigation, conversation trees, clue categories, and the four-step accusation flow.

I built a plugin that exports IntelliJ run configs as shell scripts, Docker commands, and Makefiles. Stack Overflow said it couldn't be done. Now it can.

I run 6-10 Claude sessions at once. Dispatch and remote control are single-session tools. So I built a mobile PWA that wires into IntelliJ via Terminal Bridge and Tailscale — here's the architecture.

I built four IntelliJ plugins that give AI agents full IDE access — terminals, builds, notifications, project intelligence. Here's what happened when I stopped pasting Maven commands and started letting agents operate the IDE directly.