Cord: Coordinating Trees of AI Agents
Summary
Cord introduces a runtime for coordinating trees of AI agents using spawn, fork, and ask primitives to create dynamic, context-aware task trees at runtime. It critiques fixed-coordination models and demonstrates how agents can decompose work with dependencies and parallelism, while incorporating human input for scale-sensitive decisions. The piece reports tests showing reliable task decomposition and context propagation, and provides instructions to try Cord via GitHub.