Proxify Talks: Self-Improving Agents: From Prompt Engineering to Hyperagents

💡 Self-Improving Agents: From Prompt Engineering to Hyperagents As foundation models become increasingly commoditized, much of an AI system’s performance depends on the surrounding agent harness: its prompts, tools, memory, retrieval, workflows, and control logic. At this Proxify Talks session, Lele Cao - Senior Principal AI/ML Researcher at King (a part of Microsoft) will trace how optimization has evolved from prompt engineering to agents that can improve not only their own harnesses, but also the improvement process itself. Lele will introduce recent ideas such as meta-harnesses and hyperagents and discuss why open-ended search and transferable self-improvement matter. Join us and learn:⚡ Why the agent harness matters as much as the foundation model⚡ How AI systems can improve their own harnesses - and the improvement process itself⚡ What meta-harnesses and hyperagents mean for the future of AI⚡ Real-world examples from paper review and game playtesting If you work with AI, ML, research, or engineering, or simply want to understand where AI systems are heading next, this session is for you. ​Agenda: 📅 Tuesday, September 22nd, 2026🕠 08:15📍 Proxify HQIf you have any questions, don’t hesitate to reach out to us at [email protected]