Tag
The article distills 28 research papers into a 10-layer stack for building self-improving harnesses around AI models, emphasizing bounded, gated changes over general agent loops.