[orbital-decay]

>Both work because the priming step alone ticks a bunch of boxes the over-prompter is trying to hit one at a time. Once the agent has read the reference it already matches the conventions, the naming patterns, the way functions are structured, and how errors get handled, all the stuff people otherwise try to spell out constraint by constraint.

This only works either if you are living in the future where AGI already happened, or if you're trying to work with extremely mundane things. My experience is the complete opposite, current models produce hot garbage when trying to familiarize themselves with the codebase. That's precisely the reason I'm trying to correct them by adding manual constraints. Not vice versa. Predicting what's going to be important and what's not seems to be a major bottleneck for them. They're easily overwhelmed.