multi-agent-orchestration
When multiple agents win, when they lose to a single agent at equal token budget, and what it costs in money, verification and trust boundaries.
The subagent management bottleneck is granting privilege, not perception Can a leader model manage a team when it only sees part of the workspace? Kaiwen Xiong, Haonian Ji, Shi Qiu +4 · 7 authors · UNC-Chapel Hill, University of California, Berkeley, and University of California, Santa Cruz · 2026 · arXiv:2606.31174 Same task, same models, different topology: over 30 points and double the time Does the org chart change the build when the models stay the same? Yanyu Ren, Yunfeng Bai, Xizheng Wang +2 · 5 authors · Tsinghua University and Zhongguancun Laboratory · 2026 · arXiv:2607.27877 Replace the orchestrator with a shared queue and the swarm gets faster and more accurate What is the main agent costing you while it waits? Yuzhen Mao, Jerry Gu, Aadi Chauhan +3 · 6 authors · Stanford University · 2026 · arXiv:2606.10662 Delegation without trust: what a hijacked sub-agent still must not be allowed to do If a sub-agent can be hijacked, what is it still not allowed to do? Panduranga Sai Varma Dantuluri, Jyotirmoy Sundi · 2 authors · VotalAI · 2026 · arXiv:2609.00267 In agent teams the scarce resource is reading, not message delivery What should an agent read when its context is already full? Karthik Sreedhar, Leo Yuan, Pradyumna Chari +1 · 4 authors · MIT Media Lab · 2026 · arXiv:2610.36264 From monoliths to swarms: role decomposition brings attack surfaces a single agent never had Does splitting the work across agents open an attack a single agent cannot have? Yashaswi Malla, Sandra Siby · 2 authors · New York University Abu Dhabi · 2026 · arXiv:2608.00202 Count the tokens first: single agents hold up against swarms on multi-hop reasoning Does the swarm still win once you count every thinking token? Dat Tran, Douwe Kiela · 2 authors · Stanford University · 2026 · arXiv:2604.02460 Subagents vs agent skills: when a reusable package should run in its own context Should a skill run inside your context, or in a child’s? Wasu Top Piriyakulkij, Rachel Lawrence, Alicia Curth +2 · 5 authors · Cornell University and Microsoft Research Cambridge · 2026 · arXiv:2609.09233 When the child inherits: what a spawned sub-agent carries over from its parent What exactly does a sub-agent inherit when you spawn it? Ziwen Cai, Yihe Zhang, Xiali Hei · 3 authors · University of Louisiana at Lafayette · 2026 · arXiv:2605.08460 At what network size do frontier models stop coordinating? At what network size does adding an agent stop helping? Florian Grötschla, Luis Müller, Jan Tönshoff +2 · 5 authors · ETH Zurich, RWTH Aachen University, and Google Research · 2025 · arXiv:2507.08616 Budget first, agents second: structuring multi-agent systems under a hard cost ceiling What would your agent team look like if the budget came first? Liming Yang, Junyu Luo, Xuanzhe Liu +2 · 5 authors · Peking University, University of Illinois Urbana-Champaign, Nanyang Technological University, and Tsinghua University · 2025 · arXiv:2511.21572 MASS searches role prompts and agent wiring as one design problem Do you tune what each agent says, or how the agents are wired? Han Zhou, Xingchen Wan, Ruoxi Sun +5 · 8 authors · Google and University of Cambridge · 2025 · arXiv:2502.02533 Scale the number of verifiers, not the size of the model What if reliability came from more judges instead of a better model? Shalev Lifshitz, Sheila A. McIlraith, Yilun Du · 3 authors · ArdaLabs.AI, University of Toronto, Vector Institute for AI, and Harvard University · 2025 · arXiv:2502.20379 Multi-turn orchestration matches the best single model, and the voting UI changes the answer Does asking four models to agree beat asking the best one? Aaron Xuxiang Tian, Ruofan Zhang, Jiayao Tang +12 · 15 authors · Independent Researcher, Arizona State University, and University of Pennsylvania · 2025 · arXiv:2509.23537 Debate is test-time scaling, and it only pays under specific conditions When does multi-agent debate actually beat one agent thinking harder? Yongjin Yang, Euiin Yi, Jongwoo Ko +3 · 6 authors · KAIST AI, MPI for Intelligent Systems, University of Toronto, and Vector Institute · 2025 · arXiv:2505.22960
No paper matches this filter.