Agents
Routed Graph Handoff: Adaptive Format Selection for Multi-Agent LLM Delegation
arXiv:2608.25277v1 Announce Type: new Abstract: Multi-agent LLM systems coordinate through natural-language messages that consume 40--60% of their token budget. Replacing these with structured graphs
arXiv:2608.25277v1 Announce Type: new Abstract: Multi-agent LLM systems coordinate through natural-language messages that consume 40--60% of their token budget. Replacing these with structured graphs reduces cost but fails on tasks requiring adaptive reasoning. We propose extbf{Routed Graph Handoff}, where a lightweight LLM router (155 tokens, 0.15% overhead) selects between a typed dependency graph and natural language for each delegation. On four benchmarks (1,050+ trajectories), the routed system matches or exceeds NL-only on every task: extbf{+12.7,pp} on au-retail at 3.2imes compression (p{<}0.01), extbf{+8.7,pp} on BrowseComp at 2.2imes compression (p{<}0.05), and parity on BFCL and AppWorld. Without the router, graph-only delegation regresses 14.6,pp on AppWorld; the router eliminates this at near-zero cost. A graph-aware executor prompt is required: the same schema without interpretation guidance yields no gain. An oracle analysis reveals 8.6,pp of additional headroom, motivating execution-time adaptive routing as future work.
Related
- MASFactory: A Graph-centric Framework for Orchestrating LLM-Based Multi-Agent Systems with Vibe Graphing
- Roles with Rails: Contract-Preserving Role Evolution in Multi-Agent Structured Reasoning
- Focus Is All You Need: Adaptive Goal-aware Attention Orchestration for Multi-Agent Graph Systems
Source: arXiv cs.CL | 2026-08-27