AutoRAS: Learning Robust Agentic Systems with Primitive Representations

2026-06-19

AI Summary

This work addresses the widespread neglect of robustness in existing automated multi-agent system designs, which often fail under external attacks or internal faults. To bridge this gap, we propose AutoRAS, a novel framework that intrinsically embeds robustness into the design pipeline. AutoRAS unifies the representation of agent structural connectivity and behavioral actions through symbolic primitives and leverages safety signals from execution feedback alongside streaming sequence-level optimization objectives to enable end-to-end generation of robust systems. Empirical evaluations demonstrate that AutoRAS achieves state-of-the-art performance in both standard and adversarial environments, exhibits minimal performance degradation under attack, and displays strong transferability, optimization stability, resilience to primitive-set variations, and favorable cost-effectiveness.

Abstract

The automated design of agentic systems offers a promising pathway for scaling large language models (LLMs) beyond single-agent reasoning. While prior work has advanced task performance through handcrafted or automatically generated multi-agent workflows, robustness is often treated as an afterthought, leaving systems vulnerable to external adversaries and internal failures. We propose AutoRAS, a framework for the Automated design of Robust Agentic Systems. AutoRAS formulates system design as generating a sequence of symbolic primitives that jointly encode structural connectivity and behavioral actions, and learns to optimize this sequence using execution-derived safety signals and flow-based sequence-level objectives. Extensive experiments show that AutoRAS achieves the best performance in both vanilla and adversarial settings, with the smallest performance degradation under attacks. Further analyses demonstrate strong transferability, stable optimization behavior, stability across primitive sets, and favorable cost trade-offs. Our code is available at $\href{https://github.com/guohezuy/AutoRAS}{\text{this https URL}}$.

Problem

• robustness
• agentic systems
• adversarial attacks
• system failure
• automated design

Innovation

• AutoRAS
• robust agentic systems
• symbolic primitives
• flow-based optimization
• adversarial robustness