dispatching-parallel-agents

Dispatch parallel agents to handle independent tasks simultaneously. Identify 3+ distinct problems, assign each agent a focused scope with clear goals and constraints, and run them concurrently via the Task tool. Results integrate after all agents complete, avoiding shared-state conflicts.

Parallelize when you have 3+ independent tasks where each can succeed or fail without affecting others. Use parallel agents for separate domains—test clusters, feature areas, or subsystems—where work isolation prevents interference and integration happens after completion.

Group tasks by independence first, then create focused agent tasks with explicit scope, goal, constraints, and context. Each agent works within its domain boundaries. Follow a four-step pattern: group, create scoped tasks, dispatch concurrently, and integrate results with review steps between domains.

Parallel agents work best for independent tasks without shared state. If tasks share dependencies, they cannot run safely concurrently. Verify independence before dispatching—each agent must operate within its own scope without requiring outputs from other parallel agents.

Collect summaries from each parallel agent, then perform integration and review steps across domains. The system handles concurrent execution; you define integration logic to merge results from independent agents into a unified solution.