So, capabilities/type systems. Building code architecture guardrails steep enough the AI won't jump the fence/take shortcuts.

I think it's all about keeping state in the determinant space. I've come across the same issue, the key was to not rely on LLM performing workflow - the runtime needs to enforce.

Author here. The TL;DR: move rules from prompts into types the compiler refuses to violate, then bounce the AI coding loop off those refusals. The repo is github.com/pyrex41/Shen-Backpressure. Builds a lot on Geoff Huntley's backpressure idea -- none of this is rocket science, just an effort to apply sound programming principles in a world of LLM coding agents.