what's new in ai agents & automation

Ismael Rousseau, Geraldine Damnati, Frederic Bechet

this paper focuses on stress-testing deep search agents to evaluate their robustness against poor-quality evidence. practitioners can use these insights to build more resilient search agents that perform reliably even with imperfect information, which is common in real-world data.

Minghao Liu, Yu Wang, Jiayun Wang +1

this paper introduces a method for self-evolving agents without explicit rewards, using a pairwise validator. this is valuable for practitioners developing agents in environments where reward functions are hard to define, enabling continuous improvement through self-correction.

Xinyu Geng, Xuanhua He, Sixiang Chen +7

this research introduces a self-distillation method for deep search agents operating in verifiable environments. it aims to improve the efficiency and effectiveness of agents in complex search tasks. practitioners can leverage this to develop more capable and robust search-oriented ai agents.

Hongliang Li, Yijin Liu, Zhiwei Zhang +5

this paper introduces a benchmark, polyworkbench, for evaluating multilingual llm agents on long-horizon tasks. it helps assess agent performance across different languages and complex, multi-step problems. practitioners can use this benchmark to compare and improve the capabilities of their llm agents.

Laïla Elkoussy, Julien Perez

this paper presents a framework for verifying and enforcing procedural compliance in llm agents that use tools. it helps ensure agents follow specified rules and procedures, which is crucial for reliable and safe deployment. practitioners can use this to build more trustworthy and controllable ai agents.

Yuchen Xia

this paper introduces a system that integrates large language model agents with digital twins for industrial automation. it aims to enhance the autonomy of cyber-physical systems in production environments, allowing practitioners to build more intelligent and adaptive industrial control systems.

Petar Radanliev, Omar Santos, Carsten Maple +1

this paper proposes a framework for advisory automation in agentic ai using aibom and csaf-vex. it helps practitioners ensure the security and trustworthiness of ai agents by providing a reproducible method for managing and verifying their components and behaviors.

this paper proposes memgui-agent, an end-to-end mobile gui agent designed to handle long-horizon tasks by proactively managing context. this is highly relevant for developers creating automation tools for mobile applications, as it addresses the challenge of agents reliably performing multi-step operations over time.

this research discusses the development of autonomous medical ai agents that can integrate into clinical workflows and access patient data. this is crucial for healthcare practitioners and developers aiming to build physician copilots that go beyond simple chat tools and perform actions within ehr systems.

this paper presents an llm-based ai agent designed for deploying virtual network functions, moving towards more intelligent network management. practitioners in network operations and cloud infrastructure can use this to explore automating complex deployment tasks with ai.

Drewry H. Morris V, Luis Valles, Reza Hosseini Ghomi

graphflow introduces a visual workflow system that enables formally verifiable agentic ai automation. this is important for practitioners who need to build reliable and trustworthy ai agents, especially in critical applications where correctness and safety are paramount. it helps ensure that automated processes behave as expected.

Kishan Athrey, Ramin Pishehvar, Brian Riordan +1

proposes methods to automatically compose multi-agent systems by recommending which agents should handle specific user intents. helps practitioners build flexible, reusable agent families without manual orchestration.