Recursive self-improvement of AI research agents.

arXiv:2609.26457v1 Announce Type: new Abstract: AI agents are beginning to automate research and development across the AI stack, from improving training efficiency to optimizing inference. A natural next step is to improve the research efficiency of the agent