Sub-Agents & Parallel Execution - Which condition makes it safe to dispatch multiple sub-agents in parallel rather than sequentially?

The sub-tasks are independent — no sub-agent needs the output of another to complete its work

All sub-agents use the same set of tools

The total token budget is large enough to hold all contexts simultaneously

The coordinator has already completed at least one sequential run to validate the approach