Training-Free Agentic AI: Probabilistic Control and Coordination in Multi-Agent LLM Systems
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2026
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866911514485587968 |
|---|---|
| author | Hosseini, Mohammad Parsa Shah, Ankit Qureshi, Saiyra Huang, Alex Miao, Connie Wei, Wei |
| author_facet | Hosseini, Mohammad Parsa Shah, Ankit Qureshi, Saiyra Huang, Alex Miao, Connie Wei, Wei |
| contents | Multi-agent large language model (LLM) systems enable complex, long-horizon reasoning by composing specialized agents, but practical deployment remains hindered by inefficient routing, noisy feedback, and high interaction cost. We introduce REDEREF, a lightweight and training-free controller for multi-agent LLM collaboration that improves routing efficiency during recursive delegation. REDEREF integrates (i) belief-guided delegation via Thompson sampling to prioritize agents with historically positive marginal contributions, (ii) reflection-driven re-routing using a calibrated LLM or programmatic judge, (iii) evidence-based selection rather than output averaging, and (iv) memory-aware priors to reduce cold-start inefficiency. Across multi-agent split-knowledge tasks, we show that while recursive retry alone saturates task success, belief-guided routing reduces token usage by 28%, agent calls by 17%, and time-to-success by 19% compared to random recursive delegation, and adapts gracefully under agent or judge degradation. These results demonstrate that simple, interpretable probabilistic control can meaningfully improve the efficiency and robustness of multi-agent LLM systems without training or fine-tuning. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_13256 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Training-Free Agentic AI: Probabilistic Control and Coordination in Multi-Agent LLM Systems Hosseini, Mohammad Parsa Shah, Ankit Qureshi, Saiyra Huang, Alex Miao, Connie Wei, Wei Computation and Language Artificial Intelligence Emerging Technologies Multiagent Systems Multi-agent large language model (LLM) systems enable complex, long-horizon reasoning by composing specialized agents, but practical deployment remains hindered by inefficient routing, noisy feedback, and high interaction cost. We introduce REDEREF, a lightweight and training-free controller for multi-agent LLM collaboration that improves routing efficiency during recursive delegation. REDEREF integrates (i) belief-guided delegation via Thompson sampling to prioritize agents with historically positive marginal contributions, (ii) reflection-driven re-routing using a calibrated LLM or programmatic judge, (iii) evidence-based selection rather than output averaging, and (iv) memory-aware priors to reduce cold-start inefficiency. Across multi-agent split-knowledge tasks, we show that while recursive retry alone saturates task success, belief-guided routing reduces token usage by 28%, agent calls by 17%, and time-to-success by 19% compared to random recursive delegation, and adapts gracefully under agent or judge degradation. These results demonstrate that simple, interpretable probabilistic control can meaningfully improve the efficiency and robustness of multi-agent LLM systems without training or fine-tuning. |
| title | Training-Free Agentic AI: Probabilistic Control and Coordination in Multi-Agent LLM Systems |
| topic | Computation and Language Artificial Intelligence Emerging Technologies Multiagent Systems |
| url | https://arxiv.org/abs/2603.13256 |