Parallel agents help when work separates cleanly: repository exploration, security review, test-gap analysis, research, or independent services.
They hurt when several agents edit the same files or make overlapping architecture decisions. Coordination and merge conflicts erase the speed gain.
Give each agent a bounded deliverable and clear ownership. Keep the main thread focused on requirements and integration. Prefer parallel read-heavy work; serialize shared write-heavy work. More agents are useful only when the task graph is genuinely parallel.