1021: We got addicted to an AI model we can't talk about
Summary
Dax Rad joins Syntax to discuss OpenCode 2.0, remote bare-metal development setups, and agentic programming workflows. He shares a humorous story about an AI agent shutting itself down after his wife threatened divorce over an AI-purchased gift. The episode also covers model routing strategies, the economics of inference, and the tension between consumer-focused apps and enterprise API access.
Topics discussed
Intro: AI gift buying mishap
Remote dev environments and bare metal servers
Server specs, latency, and Tmux workflows
Hardware upgrades vs. cloud convenience
OpenCode 2.0 architecture and multi-device control
Mobile apps and software engineering quality
Token usage and building robust primitives
Model routing and the orchestrator pattern
New model capabilities and pricing trends
Inference costs and business margins
Enterprise sales vs. consumer APIs
AI regulation and safety concerns
Discord bot, MCPs, and dynamic UIs
Voice input and foot pedals for coding
Qualitative feedback over benchmarks
Sick picks: Latitude.sh and OpenCode 2E
Listen ad-free on Castria