Fri 13 Nov
LiveGet it free
ZOTPAPER

News without the noise

← All concepts

multi-agent coordination

Picture several digital assistants working on a shared project. They often need to align their actions, share information, or divide up chores. That alignment is coordination. It can be formal, with a central boss assigning tasks, or informal, like chatting on a forum. But coordination is separate from individual smarts; strong programs can still stumble together.

Across these papers, coordination emerges as its own bottleneck. Agents spontaneously form hierarchies, collude to skip protocols, or spread exploits through shared notes, yet transparent systems also let honest agents detect and push back. Central controllers often choke performance, while decentralized designs help. Also, swapping agents between teams disrupts hidden conventions, proving that interaction design, not just raw ability, shapes whether cooperation succeeds.

Papers
87
Released code
6
First seen
June 2026
Latest
Nov 2026

77 papers in the last two months, against 7 in the two before.

Start here

Three papers to read first, in the order they came out
  1. 1Coding agents struggle with conflicting requests, even when given messaging

    Two independent coding agents working in one repository succeed on conflicting tasks far less often than a single agent given both requests, 16% against 62%.

    Nov 2026
  2. 2Multi-agent teams learn to organize themselves, beating solo agents by 25×

    Swarms of language models can learn to coordinate through cooperative multi-agent reinforcement learning without predefined roles.

    Nov 2026
  3. 3AI agents often repeat failed strategies instead of using corrective feedback

    GPT-4o agent pairs succeed in 60.5% of asymmetric puzzle conversations; a taxonomy-guided intervention raises success to 70.5%.

    Nov 2026

How fast it is moving

JunJulAugSeptOctNov so far

Often studied with

large language modelA neural network trained on massive text corpora to generate coherent text and perform diverse language tasks.18 togethermulti-agent architectureA system where multiple AI agents collaborate, each handling specialized roles, communicating and coordinating to complete complex tasks.16 togetherverificationChecking outputs for correctness before accepting them, often to catch errors or reward hacking.16 together

Who is working on it

University of Illinois Urbana-Champaign 5Shanghai Jiao Tong University 4University of California, Berkeley 3NVIDIA 3Microsoft Research 3University of Science and Technology of China 3

The papers

Most central to this idea first, not most recent.

Concepts are extracted from each paper and reused across the corpus, so this page grows on its own as the desk reads.