Survey Maps Self-Improvement in Modern Agentic AI Systems.

Autonomous agents capable of self-improvement are transitioning from theoretical concepts to practical deployments. The core objective of these systems is to adapt and evolve based on experience, ideally with minimal or no human intervention. This survey introduces a comprehensive system-level framework to understand these modern self-improving agents. Within this framework, an agent is conceptualized as a foundation model coupled with an operational scaffold, which includes prompts, memory, tools, and control logic. Self-improvement is formally defined as an operator that autonomously generates and applies updates to either the model's parameters or its scaffold components. The survey categorizes existing research based on what is being updated and the signals that trigger these changes, concluding with a discussion of applications, evaluation methods, and future research directions in this rapidly evolving field.