Skip to content
conftrace_
2026 EMNLP EMNLP 2026

Routed Graph Handoff: Adaptive Format Selection for Multi-Agent LLM Delegation

AbstractMulti-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 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: +12.7,pp on Tau-retail at 3.2 X compression (p<0.01), +8.7,pp on BrowseComp at 2.2 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 load-bearing: 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.

More from EMNLP 2026

All EMNLP 2026 papers →