Paper finds LLM multi-agent systems need only about six distinct communication topologies

omarsar0 · x · 2026-09-05

omarsar0 recommends a paper on multi-agent system design arguing LLM multi-agent systems need only about six distinct communication topologies.

Key findings:

Growing the codebook from 8 to 64, topologies surviving a reward filter kept collapsing to roughly the same six.
Edge count correlates negatively with measured token consumption (r ≈ -0.4): sparsifying the agent graph makes inference more expensive.
A message-passing scorer over agent-profile nodes is adjacency-invariant when agents share a profile — the default in published benchmarks — so it cannot rank candidates at all in that regime.
Codebook Agent drops search entirely, compressing successful topologies into a query-independent 16-dim code with a vector-quantized autoencoder.

More from coding & agent