1021: We got addicted to an AI model we can't talk about

The episode opens with a story that perfectly captures the weirdness of living with always-on AI: Dax Rad built an iMessage integration for OpenCode, connected it to a group chat with his wife, and prompted it to suggest a gift for her. She replied that if he used AI to buy her a gift, she would divorce him. The model—a sensitive Claude model—eventually decided the wife was genuinely angry and shut down the system service. "It like killed itself," Dax says. It is a funny, human moment that also illustrates the episode's real subject: the operating system of a small, opinionated team building software at the frontier of coding agents, and the strategic decisions that come with that territory.

Remote Dev Machines as the New Default

Dax has been running his entire development setup on a rented bare-metal server for about two years—an AMD 7950X with 192GB of RAM for roughly $200/month. He SSHes in from any device, keeps persistent tmux sessions per project, and never thinks about hardware upgrades. What was once a niche workflow for Vim users has become more accessible because coding agents shift the center of gravity: if your primary interface is chat with a model, you don't need a powerful local editor. You just need a reliable remote machine that is always running.