ADIAS: New Framework for Automated Agent Design via Issue-Centric Optimization

A recent study presents ADIAS (Automated Design of Interactive Agentic Systems), a new framework aimed at automating the design of full-code agents. This research, which can be found on arXiv (ID 2608.06410), highlights the shortcomings of current automated agent design approaches that focus on candidates and structure cross-round experiences around them, resulting in implicit repair progress. Such a method leads to ineffective targeting of repairs, slow consolidation of advancements, and the continuation of unproductive interventions. To address these challenges, the authors propose an issue-centric optimization strategy, where repair progress is explicitly maintained as a persistent issue state. ADIAS employs two key mechanisms: a stable issue state that tracks identities, lifecycle status, evidence, and intervention histories, and issue-guided optimization to suggest repair targets. This preprint, authored by unnamed researchers, aims to enhance the efficiency and effectiveness of automated agent design, which is crucial for advancing AI systems and interactive agentic systems.