ADIAS: Automated Design of Interactive Agentic Systems

2026-08-03

AI Summary

Existing agent-based optimization methods are candidate-centric, leading to ambiguous repair objectives, inefficient integration of progress, and the accumulation of ineffective interventions. This work proposes the first problem-centric optimization framework, which explicitly models and persistently maintains problem states—including identity, lifecycle, evidence, and intervention history—to uniformly guide code-level repair targets and modification directions. By departing from conventional paradigms, the approach achieves an average improvement of 25.2% over the strongest baseline across five interactive benchmarks and consistently enhances performance across four backbone models. Ablation studies further demonstrate that omitting key components can degrade performance by up to 40.7%, underscoring their critical role in the framework’s efficacy.

Abstract

Automated agent design improves agent harnesses through iterative revision, evaluation, and feedback summarization. Existing methods are largely candidate-centric: cross-round experience is organized around candidate agents, which leaves the repair progress implicit. This causes inefficient repair targeting, slow consolidation of partial progress, and propagation of ineffective interventions across rounds. Therefore, we formulate issue-centric agent optimization, in which repair progress is carried forward as an explicit persistent issue state to guide optimization, rather than re-derived from candidate history in each round. We instantiate the formulation in ADIAS, a framework for automated full-code agent design with two mechanisms. A persistent issue state maintains stable issue identities, lifecycle status, supporting evidence, and intervention-outcome histories. Issue-guided optimization uses this state to jointly propose repair targets and revision directions for subsequent focused full-code modification. Across five interactive benchmarks, ADIAS outperforms the strongest baseline by 25.2% on average and achieves consistent gains across four backbone models. Controlled ablations further show that removing persistent issue state or replacing issue-centric revision with candidate-centric policies leads to performance drops of up to 40.7%.

Problem

• automated agent design
• candidate-centric optimization
• repair progress
• interactive agentic systems
• issue-centric optimization

Innovation

• issue-centric optimization
• persistent issue state
• automated agent design
• interactive agentic systems
• full-code revision