SAGE: A Statistical Acceptance Gate for Self-Evolving Agents

A new statistical framework to govern autonomous agent self-improvement and prevent logic degradation.

Large Language Model (LLM)-based agents increasingly self-evolve by editing a persistent skill document that encodes their workflow, tool-use rules, and decision logic. This loop has two steps, an optimizer that proposes a candidate edit and a gate that accepts or rejects it. Pri...

ArXiv CS.AI2026-09-30