According to God of Prompt on X, Meta AI introduced Hyperagents, a framework where a task agent and a meta agent are unified so the system can modify both agents and the modification process itself, enabling metacognitive self-modification and compounding improvements across domains (as reported by the cited tweet). According to the same source, Hyperagents delivers continuous gains in coding, paper review, robotics reward design, and Olympiad-level math grading, outperforming baselines without self-improvement and prior systems such as the Darwin Gödel Machine. As reported by the post, the key advance is that improvements to the improvement process—such as persistent memory and performance tracking—transfer across domains and accumulate over runs, addressing a fundamental limitation of earlier self-improving systems that were domain-locked to coding. For AI builders, this suggests new business opportunities in automated agentic pipelines, cross-domain evaluation tooling, and enterprise copilots that learn how to optimize themselves over time, according to the X thread’s summary of the paper.