Release Recap: Scaling Inference-Time Compute with Teams of Coding Agents

​Join Keenan Samway, Daniel D'souza, and Nouha Dziri as they explore a novel approach to multi-agent system design for scaling inference by framing team construction as a compute allocation problem. They demonstrate that strategic delegation—specifically assigning workers to planning rather than implementation or verification—can achieve near-optimal reward at a fraction of the cost, with a heterogeneous 17-agent team reaching within 1.4 points of a strong homogeneous 5-agent team at 70% lower cost, thereby offering actionable guidelines for optimizing agent budgets under task, cost, and latency constraints.

​Read the paper: https://
cohere.link/tStRYJS
Check out the blog post: https://
cohere.link/0TzgKoy