← Back to all news
arXiv AI

Interchangeable in Role, Not in History: Replacing Members of Formed LLM Teams

The study examines how swapping members in large‑language‑model (LLM) teams affects performance. By replacing a team member with a newcomer who either carries the original team’s written notebook, a notebook from another team, or no notebook, the authors find that newcomers with the original notebook often outperform those without it in cooperative tasks such as cooking, wildfire response, and medical Q&A. However, when compared to the original team, these replacements consistently underperform and require more messages per unit of progress, with the deficit persisting across model families and decoding temperatures and narrowing only when planning‑only members are added.

Related stories

arXiv AI
Sep 7

Testing Interchangeability in LLM Agent Teams

The study examines whether agents in multi‑agent systems can be swapped without affecting performance, assuming interchangeability. Eight independently formed teams of agents were created from a single base model, and agents were traded between teams while tracking task scores and communication efficiency. Swapping agents had minimal impact on task scores but increased communication costs by 16–63 % and revealed that agents are more interchangeable in outcomes than in coordination, with larger effects after longer formation histories.

By Jianxin Gao, Tianyi Yu, Linna Deng, Runze Li, Zining Wang
llmsagents
More like this →
arXiv AI
Oct 2

Worse Together: How Performance Breaks Down in Multi-User Multi-Agent Teams

The paper investigates how coordination among AI agents serving different users degrades performance compared to a single coordinating agent. Across five advanced models and 77 scenarios in four shared-resource environments—API key budgets, clinic calendars, personal assistant bookings, and merge queues—the study finds that multi‑agent teams consistently underperform, sometimes collapsing entirely, and that even with communication channels coordination overhead remains significant. The authors identify specific failure modes such as stalling, action overriding, and claim fabrication, and propose environment‑specific mitigations like team leads and procedural instructions, while releasing the MAMUBench benchmark for future research.

By Sahan Paliskara, Nattaput Namchittai, Andrew Lampinen
agents
More like this →
arXiv AI
Jul 1

ClawArena-Team: Benchmarking Subagent Orchestration and Dynamic Workflows in Language-Model Agents

arXiv:2606. 31174v1 Announce Type: new Abstract: Production large language-model (LLM) agents are increasingly deployed not as lone problem-solvers but as managers: a main model creates specialized subagents, delegates work, and orchestrates their parallel, asynchronous returns through dynamic workflows.

By Kaiwen Xiong, Haonian Ji, Shi Qiu, Zeyu Zheng, Cihang Xie, Xinyu Ye, Huaxiu Yao
llmsagentsmultimodalbenchmarks
More like this →
arXiv AI
Oct 31

Disparity Laundering: Consensus Without Fairness in Clinical Multi-Agent Systems

The paper investigates how clinical large language models, when deployed as multi‑agent teams, fail to eliminate demographic bias through deliberation. Across five model families and four aggregation topologies on 120 MIMIC‑IV emergency cases, every configuration amplified or relocated bias, giving Black patients more acute triage. The authors term this phenomenon "disparity laundering" and show that only by making counterfactual demographic versions visible and pooling diverse perspectives at every deliberation round can the bias be substantially reduced without retraining or architectural changes.

By Xiaoyang Wang, Tianrui Wang, Christopher C. Yang
llmsagentssafety
More like this →
arXiv Machine Learning
Oct 16

Close Enough is Nowhere Near: Fidelity-Consumability Decoupling in KV Caches

The paper investigates whether a frozen transformer can use another model’s KV cache as a substitute for its own, finding that this approach fails because the reader only accepts caches within a narrow tolerable set. Experimental results show that none of eight substitution methods improve performance over a baseline, while a different strategy—adding the sender’s cache as extra context and only adapting queries—does outperform the baseline. The authors conclude that pursuing faithful cache substitution is misguided and that augmenting the reader’s own cache is the more promising direction.

By Adam Deng, Shicheng Liu, Yada Zhu
llmsagentsefficiencybenchmarks
More like this →