Use when subagent capacity is available and two or more independent problem domains can run at the same time on non-overlapping files without shared state, such as separate root-cause investigations in different components or dependency-ready tasks with distinct write scopes.
