Use when a question or task is broad enough that one pass will miss things - research, audits, reviews, comparisons, migrations, "be thorough" requests - and the work can be split across parallel agents whose results must be verified in fresh context and merged into a single report.
