Hu, S., Lu, C., & Clune, J. (2024). Automated Design of Agentic Systems. arXiv.

The paper titled “Automated Design of Agentic Systems” by Shengran Hu, Cong Lu, and Jeff Clune (2024) introduces a novel research area termed Automated Design of Agentic Systems (ADAS). The paper discusses how traditional hand-designed solutions in AI are increasingly being replaced by learned solutions, proposing that agentic systems can be automatically created, including inventing new building blocks or combining existing ones in novel ways. The authors present an algorithm called Meta Agent Search, which iteratively programs and evaluates new agents based on an ever-growing archive of previous discoveries. Through extensive experiments across domains such as coding, science, and mathematics, the algorithm demonstrates its ability to invent agents with novel designs that outperform state-of-the-art hand-designed agents. The paper emphasizes the robustness and generality of the agents, as they maintain superior performance even when transferred across different domains. The work highlights the potential for ADAS to revolutionize the development of agentic systems by automating their design and suggests that this approach could lead to more efficient and effective AI solutions.

https://arxiv.org/abs/2408.08435