What question did this study set out to answer?

Identify the optimal fleet size and configuration of multi-agent LLM systems to minimize coordination overhead.

synapse

⌘+K

synapse

⌘+K

April 15, 2026Open Access

The Agent Coordination Bound (ACB): A Framework for Multi-LLM Fleet Sizing, Topology Selection, and Deployment Diagnostics

Key Points

Identify the optimal fleet size and configuration of multi-agent LLM systems to minimize coordination overhead.
Developed a structural heuristic named Agent Coordination Bound (ACB).
Derived key analytical contributions including degradation probability and topology crossover theorem.
Introduced Coordination Bottleneck Index (CBI) as a diagnostic tool.
Established that the optimal fleet size is n* = ⌈a/c⌉.
Showed that communication channels increase as n(n-1)/2.
Demonstrated that coordination overhead can outweigh reasoning benefits.

Abstract

Multi-agent LLM frameworks make it trivially easy to scale agent count, but provide no principled criterion for when to stop. This paper addresses that gap through the Agent Coordination Bound (ACB): a structural heuristic showing that inter-agent communication channels grow as n(n-1)/2 and coordination overhead eventually dominates reasoning gain. The model predicts an optimal fleet size of n* = ⌈a/c⌉. This paper derives six analytical contributions, including a degradation probability P(harm|n), a topology crossover theorem, and a Coordination Bottleneck Index (CBI) as a deployable scalar diagnostic.

The Agent Coordination Bound (ACB): A Framework for Multi-LLM Fleet Sizing, Topology Selection, and Deployment Diagnostics

Key Points

Abstract

Cite This Study