Abstract
The rapid progress of Large Language Models has advanced agentic systems indecision-making, coordination, and task execution. Yet, existing agentic systemgeneration frameworks lack full autonomy, missing from-scratch agentgeneration, self-optimizing agent functionality, and collaboration, limitingadaptability and scalability. We propose SwarmAgentic, a framework for fullyautomated agentic system generation that constructs agentic systems fromscratch and jointly optimizes agent functionality and collaboration asinterdependent components through language-driven exploration. To enableefficient search over system-level structures, SwarmAgentic maintains apopulation of candidate systems and evolves them via feedback-guided updates,drawing inspiration from Particle Swarm Optimization (PSO). We evaluate ourmethod on six real-world, open-ended, and exploratory tasks involvinghigh-level planning, system-level coordination, and creative reasoning. Givenonly a task description and an objective function, SwarmAgentic outperforms allbaselines, achieving a +261.8% relative improvement over ADAS on theTravelPlanner benchmark, highlighting the effectiveness of full automation instructurally unconstrained tasks. This framework marks a significant steptoward scalable and autonomous agentic system design, bridging swarmintelligence with fully automated system multi-agent generation. Our code ispublicly released at https://yaoz720.github.io/SwarmAgentic/.