How to Train a Swarm

Published 3 November 2026

Collective inference-time scaling enables multiple agents to perform useful work in parallel while managing communication costs. The post also outlines incentive design for parallel work and reinforcement learning.