Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Beyond Rule-Based Workflows: An Information-Flow-Orchestrated Multi-Agents Paradigm via Agent-to-Agent Communication from CORAL

Xinxing Ren, Quagmire Zang, Caelum Forder, Suman Deb, Ahsen Tahir, Roman J. Georgio, Peter Carroll, Zekun Guo

TL;DR

The paper tackles the brittleness and manual overhead of predefined workflow-based multi-agent systems by introducing an information-flow–orchestrated paradigm where a central orchestrator coordinates agents through natural-language A2A communication from CORAL. This approach eliminates fixed task-state trees and enables dynamic adjustment of instructions to handle edge cases, proven on the GAIA benchmark against a workflow-based baseline (OWL). Results show parity or superiority in pass@1 accuracy across settings (64.24% with all Grok 4.1 Fast, and 63.64% in a heterogeneous setup) with competitive token usage, driven by emergent coordination patterns and edge-case handling strategies. The work highlights four coordination patterns and three edge-case strategies arising from adaptive orchestrator–agent interactions, suggesting practical benefits for general-purpose task solving and robust monitoring in MAS. It also outlines future work to test domain-specific tasks and expand the analysis of emergent phenomena, along with detailed toolkit specifications.

Abstract

Most existing Large Language Model (LLM)-based Multi-Agent Systems (MAS) rely on predefined workflows, where human engineers enumerate task states in advance and specify routing rules and contextual injections accordingly. Such workflow-driven designs are essentially rule-based decision trees, which suffer from two fundamental limitations: they require substantial manual effort to anticipate and encode possible task states, and they cannot exhaustively cover the state space of complex real-world tasks. To address these issues, we propose an Information-Flow-Orchestrated Multi-Agent Paradigm via Agent-to-Agent (A2A) Communication from CORAL, in which a dedicated information flow orchestrator continuously monitors task progress and dynamically coordinates other agents through the A2A toolkit using natural language, without relying on predefined workflows. We evaluate our approach on the general-purpose benchmark GAIA, using the representative workflow-based MAS OWL as the baseline while controlling for agent roles and underlying models. Under the pass@1 setting, our method achieves 63.64% accuracy, outperforming OWL's 55.15% by 8.49 percentage points with comparable token consumption. Further case-level analysis shows that our paradigm enables more flexible task monitoring and more robust handling of edge cases. Our implementation is publicly available at: https://github.com/Coral-Protocol/Beyond-Rule-Based-Workflows

Beyond Rule-Based Workflows: An Information-Flow-Orchestrated Multi-Agents Paradigm via Agent-to-Agent Communication from CORAL

TL;DR

The paper tackles the brittleness and manual overhead of predefined workflow-based multi-agent systems by introducing an information-flow–orchestrated paradigm where a central orchestrator coordinates agents through natural-language A2A communication from CORAL. This approach eliminates fixed task-state trees and enables dynamic adjustment of instructions to handle edge cases, proven on the GAIA benchmark against a workflow-based baseline (OWL). Results show parity or superiority in pass@1 accuracy across settings (64.24% with all Grok 4.1 Fast, and 63.64% in a heterogeneous setup) with competitive token usage, driven by emergent coordination patterns and edge-case handling strategies. The work highlights four coordination patterns and three edge-case strategies arising from adaptive orchestrator–agent interactions, suggesting practical benefits for general-purpose task solving and robust monitoring in MAS. It also outlines future work to test domain-specific tasks and expand the analysis of emergent phenomena, along with detailed toolkit specifications.

Abstract

Most existing Large Language Model (LLM)-based Multi-Agent Systems (MAS) rely on predefined workflows, where human engineers enumerate task states in advance and specify routing rules and contextual injections accordingly. Such workflow-driven designs are essentially rule-based decision trees, which suffer from two fundamental limitations: they require substantial manual effort to anticipate and encode possible task states, and they cannot exhaustively cover the state space of complex real-world tasks. To address these issues, we propose an Information-Flow-Orchestrated Multi-Agent Paradigm via Agent-to-Agent (A2A) Communication from CORAL, in which a dedicated information flow orchestrator continuously monitors task progress and dynamically coordinates other agents through the A2A toolkit using natural language, without relying on predefined workflows. We evaluate our approach on the general-purpose benchmark GAIA, using the representative workflow-based MAS OWL as the baseline while controlling for agent roles and underlying models. Under the pass@1 setting, our method achieves 63.64% accuracy, outperforming OWL's 55.15% by 8.49 percentage points with comparable token consumption. Further case-level analysis shows that our paradigm enables more flexible task monitoring and more robust handling of edge cases. Our implementation is publicly available at: https://github.com/Coral-Protocol/Beyond-Rule-Based-Workflows
Paper Structure (13 sections, 15 equations, 5 figures, 3 tables)

This paper contains 13 sections, 15 equations, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work and Preliminaries
  3. Information-Flow-Centric Multi-Agent Coordination
  4. Evaluation and Analysis
  5. Benchmark: GAIA
  6. Baseline and Experimental Settings
  7. Main Results
  8. Case-Level Analysis of Task Coordination and Handling of Edge Cases
  9. Emergent task coordination patterns.
  10. Emergent Edge Cases Handling Strategies.
  11. Conclusion and Future Work
  12. Agent Toolkit Details
  13. Agent Prompt

Figures (5)

  • Figure 1: Decision-making tree representation of the OWL architecture.
  • Figure 2: Overview of the proposed Information-Flow-Orchestrated Multi-Agent Paradigm via Agent-to-Agent (A2A) Communication. A dedicated information flow orchestrator monitors task progress and dynamically coordinates agents through natural-language A2A interactions, eliminating the need for predefined workflows.
  • Figure 3: Cumulative distribution functions (CDFs) of token consumption for OWL and the proposed Information-Flow-Orchestrated MAS under different model configurations.
  • Figure 4: Case-level analysis of emergent task coordination patterns from the information flow orchestrator.
  • Figure 5: Case-level analysis of emergent edges cases handling from the information flow orchestrator.