Why Role-Based Subagents Slow Down Coding Agents: Subagent Folding + Progressive Skill Disclosure (2.66x Faster)

💡 AI Labo Benchmark Series — Experiment E05 (Multi-Agent vs. Folded Loop) This post is part of our empirical benchmark series testing AI coding agent harness patterns inside network-isolated Proxmox