By the fourth model swap I noticed that the part of the system I wasn’t swapping was the part holding the project together. Five configurations in three weeks: Opus 4.6, Opus 4.6 with the 1M context window, Cursor Cloud Agents on GPT Codex 5.3, Cloud Agents on Composer 2, and this week Opus 4.7. Some swaps were involuntary. Credits ran out. A model that had been my daily driver for two months started coming back with half-finished work and failing the easy parts of basic CI, which I wrote about in Caucus V1. Some were voluntary. I wanted to see what a new model could do that the previous one couldn’t. In a few cases it could. In a few cases the new one was worse at something different, and I swapped again. No architecture change on my end. No rewrite. The app is the app. The model is a junior engineer on a revolving contract. Same Prompt, Different Creature Here’s a specific observation. For a few weeks in April I was running the Caucus Permit Gate. Every PR had to land with a permit…
No comments yet. Log in to reply on the Fediverse. Comments will appear here.