Your cron jobs have been defaulting to the expensive API model this whole time. Not because they needed to. Because that's what was already configured, and nobody revisited it. 🧵 1/6
AsTheGeekLearns
@astgl.bsky.social
A systems engineer learning AI engineering in public. The wins. The mistakes. The “aha” moments that might save you hours.
Your cron jobs have been defaulting to the expensive API model this whole time. Not a pricing problem. A defaults problem nobody ever revisited. Two-line fix + a tiering rule that sticks: astgl.com/p/tiered-m...
After months with Claude Code, I had no idea what was in my own config anymore. So I mined all 80+ files—CLAUDE.md, memory, commands, and rules—as a dataset. 22 reusable workflows fell out. Already there; I just read them back. jmeg8r.net/4oVA3Sg
"Isn't running Claude Code expensive?" The honest answer: only if you send it everything. On my Mac Studio, an 8-hour local inference run costs ~5¢ in power. The same run on a frontier API: $15–30. Here's the exact routing table I use 🧵 1/7
