Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Ripple Effect Protocol: Coordinating Agent Populations

Ayush Chopra, Aman Sharma, Feroz Ahmad, Luca Muscariello, Vijoy Pandey, Ramesh Raskar

TL;DR

REP addresses brittle collective behavior in large, open populations of LLM-based agents by introducing a protocol-level coordination primitive: sharing lightweight sensitivities alongside decisions. The Ripple Effect Protocol decouples cognition from coordination and supports both numeric and textual sensitivity signals that are aggregated locally or via a median consensus for global alignment. Across Beer Game, Fishbanks, and Movie Coordination, REP substantially improves coordination speed, accuracy, and stability compared to traditional A2A communication, and scales gracefully to hundreds of agents while operating atop existing transport protocols. The work offers a scalable infrastructure for the Agentic Web and provides an SDK to enable open experimentation and deployment in real-world, multi-organization environments.

Abstract

Modern AI agents can exchange messages using protocols such as A2A and ACP, yet these mechanisms emphasize communication over coordination. As agent populations grow, this limitation produces brittle collective behavior, where individually smart agents converge on poor group outcomes. We introduce the Ripple Effect Protocol (REP), a coordination protocol in which agents share not only their decisions but also lightweight sensitivities - signals expressing how their choices would change if key environmental variables shifted. These sensitivities ripple through local networks, enabling groups to align faster and more stably than with agent-centric communication alone. We formalize REP's protocol specification, separating required message schemas from optional aggregation rules, and evaluate it across scenarios with varying incentives and network topologies. Benchmarks across three domains: (i) supply chain cascades (Beer Game), (ii) preference aggregation in sparse networks (Movie Scheduling), and (iii) sustainable resource allocation (Fishbanks) show that REP improves coordination accuracy and efficiency over A2A by 41 to 100%, while flexibly handling multimodal sensitivity signals from LLMs. By making coordination a protocol-level capability, REP provides scalable infrastructure for the emerging Internet of Agents

Ripple Effect Protocol: Coordinating Agent Populations

TL;DR

REP addresses brittle collective behavior in large, open populations of LLM-based agents by introducing a protocol-level coordination primitive: sharing lightweight sensitivities alongside decisions. The Ripple Effect Protocol decouples cognition from coordination and supports both numeric and textual sensitivity signals that are aggregated locally or via a median consensus for global alignment. Across Beer Game, Fishbanks, and Movie Coordination, REP substantially improves coordination speed, accuracy, and stability compared to traditional A2A communication, and scales gracefully to hundreds of agents while operating atop existing transport protocols. The work offers a scalable infrastructure for the Agentic Web and provides an SDK to enable open experimentation and deployment in real-world, multi-organization environments.

Abstract

Modern AI agents can exchange messages using protocols such as A2A and ACP, yet these mechanisms emphasize communication over coordination. As agent populations grow, this limitation produces brittle collective behavior, where individually smart agents converge on poor group outcomes. We introduce the Ripple Effect Protocol (REP), a coordination protocol in which agents share not only their decisions but also lightweight sensitivities - signals expressing how their choices would change if key environmental variables shifted. These sensitivities ripple through local networks, enabling groups to align faster and more stably than with agent-centric communication alone. We formalize REP's protocol specification, separating required message schemas from optional aggregation rules, and evaluate it across scenarios with varying incentives and network topologies. Benchmarks across three domains: (i) supply chain cascades (Beer Game), (ii) preference aggregation in sparse networks (Movie Scheduling), and (iii) sustainable resource allocation (Fishbanks) show that REP improves coordination accuracy and efficiency over A2A by 41 to 100%, while flexibly handling multimodal sensitivity signals from LLMs. By making coordination a protocol-level capability, REP provides scalable infrastructure for the emerging Internet of Agents
Paper Structure (23 sections, 3 equations, 3 figures, 1 table)

This paper contains 23 sections, 3 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Background and Related Work
  3. Protocol Stack for LLM Agents
  4. Coordination in Multi-agent LLM Systems
  5. Sensitivity Sharing in Distributed Systems
  6. Ripple Effect Protocol
  7. Workflow Overview
  8. Coordination and Aggregation
  9. Implementation Architecture
  10. Configuration Interface.
  11. Protocol Execution.
  12. Experimental Setup
  13. Supply Chain Coordination (Beer Game)
  14. Insight 1: REP mitigates bullwhip effects.
  15. Insight 2: Textual sensitivities outperform numerical gradients
  16. ...and 8 more sections

Figures (3)

  • Figure 1: Supply chain coordination outcomes over 20 rounds with a demand shock introduced at round 4, where customer demand doubles. (A) Baseline A2A shows persistent bullwhip oscillations as agents repeatedly over- and under-react to incomplete information. REP reduces total supply chain cost by 41.8%, from $7,300 (A2A) to $4,251 (B) REP rapidly stabilizes within 3–4 rounds, preventing runaway amplification. (C) REP textual aggregation outperforms numerical aggregation, achieving lower total costs by capturing richer causal reasoning.
  • Figure 2: Fishbanks simulation with 12 LLM agents over 8 seasons. A demand for fish remains constant while overharvesting leads to eventual depletion. (A) Fish population dynamics showing delayed resource collapse under REP compared to rapid decline in the A2A baseline. (B) Sustainability index with a critical viability threshold at 0.35; REP maintains the resource above this threshold 2 seasons longer than A2A. (C) Comparison of final outcomes across normalized metrics and profit margins. REP improves sustainability (+25.2%), population health (+28.9%), and coordination (+16.1%) while avoiding the financial losses (-2.5%) experienced by A2A agents. Results averaged over five trials.
  • Figure 3: REP performance under network sparsity and population scaling. (Left) REP maintains effective coordination even in sparse social networks (30% and 60% sparse), with only gradual increases in convergence rounds (4 → 6 → 9) while sustaining 70–75% consensus. Stars mark convergence points where the 70% threshold is reached. In contrast, A2A fails to achieve meaningful coordination even when fully connected, showing that sensitivity sharing is essential for global alignment. (Right) As population size grows from 5 to 200 agents, REP continues to converge reliably (3–15 rounds), while A2A requires 7–10 rounds at small sizes and fails entirely beyond 20 agents (shaded DNF region). Together, these results demonstrate REP’s robustness to both sparse connectivity and large-scale populations.