Multi-Paradigm Agent Interaction in Practice:A Systematic Analysis of Generator-Evaluator, ReAct Loop,and Adversarial Evaluation in the buddyMe Framework
The paper presents a systematic analysis of multi-paradigm agent interaction within the buddyMe framework. It explores three main interaction paradigms and establishes a processing pipeline along with an evaluation schema. The findings highlight the effectiveness of the Generator-Evaluator pre-review and the ReAct loop, while also noting the efficiency of adversarial discussions in content refinement.
- ▪The study identifies three principal agent interaction paradigms: Generator-Evaluator, ReAct Tool-Use Loops, and Memory-Augmented Interaction.
- ▪Generator-Evaluator pre-review detects requirement omissions in 20 percent of complex tasks.
- ▪The ReAct loop ensures stable subtask execution but results in around 30 percent redundant tool invocations.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.16821 (cs) [Submitted on 16 May 2026] Title:Multi-Paradigm Agent Interaction in Practice:A Systematic Analysis of Generator-Evaluator, ReAct Loop,and Adversarial Evaluation in the buddyMe Framework Authors:Xiaohua Wang, Chao Han, Kai Yu, XiaoLiang Xu, Liang Wang View a PDF of the paper titled Multi-Paradigm Agent Interaction in Practice:A Systematic Analysis of Generator-Evaluator, ReAct Loop,and Adversarial Evaluation in the buddyMe Framework, by Xiaohua Wang and 4 other authors View PDF Abstract:The rapid evolution of Large Language Model (LLM) agents has produced diverse interaction paradigms, yet few production systems integrate multiple paradigms within a unified architecture.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
