Source record

Published by arXiv cs.MA (T1 source). The original is at https://arxiv.org/abs/2608.06410.

Pipeline notes

The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.

ADIAS is a framework for automated design of interactive agentic systems that replaces candidate-centric optimization with issue-centric optimization, maintaining a persistent issue state with lifecycle status, evidence, and intervention histories to guide iterative full-code agent repair. Reports 25.2% average gain over baselines across five benchmarks.

A concrete framework with ablations showing that organizing agent repair around persistent issues rather than candidate history measurably changes outcomes — useful input for anyone building automated agent design loops.

Cited by

No citations on record.