This Monday morning, OpenAI engineer Weng Jiayi has proposed a groundbreaking method for Agentic AI: self-modification of its own code, a development confirmed by both 36Kr and eu.36kr.com. Instead of relying solely on more data and computing power, Jiayi's experiment showed an AI, Codex, repeatedly writing, running, and modifying its strategy code to fix failures. This process led Codex to achieve a perfect score in Atari Breakout and perform well in robot control simulations, suggesting AI cou