The AAMAS 2026 best paper award went to "Developing Guidelines for Human-LLM Agent Teams: A Multi-Stakeholder Lens" by Mireia Yurrita, Davide Dell’Anna, Pradeep K. Murukannaiah, Catholijn M Jonker and Pinar Yolum. The award was presented at the 25th International Conference on Autonomous Agents and Multiagent Systems, held 25-29 May in Paphos, Cyprus.
The conference gave awards in three categories: best paper, best student paper and best blue sky paper. Each had one winner and a list of nominees. The AAMAS 2026 best paper category had the longest list, with ten nominated papers alongside the winner.
The best paper category
The winning paper addresses teams that combine people with agents built on large language models. Its title points to a multi-stakeholder approach to writing guidelines for such teams. The source gives only the title and the authors, so the paper's method and findings are not described here.
Ten papers were nominated in the same category. Several deal directly with physical or embodied systems. UNCAP, by Neel P. Bhatt, Po-han Li, Kushagra Gupta, Rohan Siva, Daniel Milan, Alexander Todd Hogue, Sandeep P. Chinchali, David Fridovich-Keil, Zhangyang Wang and Ufuk Topcu, covers uncertainty-guided neurosymbolic planning that uses natural language communication for cooperative autonomous vehicles.
Longyin Cui was nominated for "Ratio-Based Signaling for Source-Victim Separation in Swarm Fault Detection." MeCo, from Baiqing Wang, Helei Cui, Bo Zhang, Xiaolong Zheng, Bin Guo and Zhiwen Yu, targets LLM-empowered multi-robot collaboration through similar task memoization.
The remaining nominees sit closer to theory and applied planning. Manuel Eberl and Patrick Lederer were nominated for "The Impossibility of Strategyproof Rank Aggregation." Seongmin Kim, Giseung Park, Woojun Kim, Jiwon Jeon, Seungyul Han and Youngchul Sung were nominated for generalized per-agent advantage estimation in multi-agent policy optimization.
Maoyuan Li, Zhongsheng Wang, Haoyuan Li and Jiamou Liu were nominated for R-Debater, a retrieval-augmented debate generation system built on argumentative memory. Benedict Russell, Chin-wing Leung and Paolo Turrini were nominated for work on learning partner selection in optional social dilemmas without prior information.
Nicolas Schwind, Gauvain Bourgne and Katsumi Inoue were nominated for characterizing consensuses in belief flow networks. Nimrod Talmon and Ehud Shapiro were nominated for "Grassroots Federation: Fair Democratic Governance at Scale." Yohai Trabelsi, Guojun Xiong, Fentabil Getnet, Stéphane Verguet and Milind Tambe were nominated for using LLMs to bring expert knowledge into algorithmic planning for health facility location in Ethiopia.
Student and blue sky awards
The Pragnesh Jay Modi Best Student Paper Award went to "Planning Ahead with RSA: Efficient Signalling in Dynamic Environments by Projecting User Awareness across Future Timesteps." The authors are Anwesha Das, John Duff, Jörg Hoffmann and Vera Demberg.
Five student papers were nominated. Patrick Becker, Alexander Fries, Matthias Greger and Erel Segal-Halevi submitted work on efficiently computing equilibria in budget-aggregation games. Grant Douglas, Stephen Franklin, Claudia Szabo and Mingyu Guo submitted "Building Large-Scale Drone Defenses from Small-Team Strategies."
Jessica Lally, Milad Kazemi and Nicola Paoletti were nominated for robust counterfactual inference in Markov decision processes. Yue Zhang, Zhe Chen, Daniel Harabor, Pierre Le Bodic and Peter J. Stuckey were nominated for flow-based task assignment in large-scale online multi-agent pickup and delivery. Siyue Ren, Wanli Fu, Xinkun Zou, Chen Shen, Yi Cai, Chu Chen, Zhen Wang and Shuyue Hu were nominated for work on reputation as a remedy for cooperation collapse in LLM-based multi-agent systems.
The Blue Sky Ideas Award went to Florent Delgrange for "Foundation World Models for Agents that Learn, Verify, and Adapt Reliably Beyond Static Environments." Two papers were nominated. One is "Guiding Sociotechnical Systems toward Value-Norm Equilibrium," by Nirav Ajmeri, Marina De Vos, Davide Dell’Anna, Pradeep K. Murukannaiah, Vivek Nallur, Luis Gustavo Nardin and Munindar P. Singh. The other is "The Dynamic Turn in Strategy Logics," by Rustam Galimullin, Maksim Gladyshev, Munyque Mittelmann and Nima Motamed.
What the lists show
Across the three categories, language models appear in the winning best paper and in several nominated papers. They show up in team guidelines, debate generation, multi-robot collaboration, health facility planning and cooperation among LLM-based agents. Dell’Anna and Murukannaiah appear as authors on both the best paper winner and a blue sky nominee.
Classical multi-agent topics also remain well represented. Rank aggregation, budget-aggregation equilibria, belief networks, strategy logics and counterfactual inference in Markov decision processes are all on the lists. The recognised work is not limited to one method family.
Analysis: what the awards do and do not tell us
The clearest pattern is that the top prize went to a paper about how people and LLM agents should work together, not to a new algorithm. That suggests the program committee sees team design and stakeholder concerns as a central open problem once language-model agents enter shared work. This reading rests on one title and its author list, so it should be held loosely.
For robotics readers, the relevant material is in the nominee lists. UNCAP, the swarm fault detection paper, MeCo, the drone defense paper and the pickup-and-delivery assignment paper all concern coordinating many machines. Two of those, UNCAP and MeCo, use natural language or LLMs as part of the coordination. That places language-based coordination of vehicles and robots among the work AAMAS reviewers rated highly, though none of these papers won its category.
Several things remain unproven from the information available. The source gives titles and authors but no results, benchmarks or deployment details. It is not possible to say how well any nominated method performs, whether it was tested on physical robots, or how it compares with earlier systems. A nomination or award signals peer assessment at one conference. It does not show that a method works outside the setting in which it was evaluated.
The guideline paper also raises a question the lists cannot answer. Guidelines for human-LLM teams are only useful if teams adopt them and if they change outcomes. Whether the winning paper offers evidence of that is not stated in the material available. Readers who want to judge the work should read the papers themselves.