parallel-execution

Parallel execution runs independent tasks concurrently in a single message by invoking multiple Task tool calls together, then aggregates results after completion. This eliminates sequential waiting and improves throughput when tasks have no data dependencies on each other.

Use parallel execution when you have multiple independent tasks with no data dependencies between them and agents are available for concurrent work. It's ideal for workloads where you need to aggregate results after all tasks complete and want to minimize total execution time.

Parallel execution requires independent tasks with no data dependencies. If tasks depend on each other's results, you must handle synchronization and result aggregation at collection points, or structure dependencies sequentially between parallel batches.

Parallel execution isolates failures so individual task failures don't block other concurrent tasks. Collect and validate all results after completion to identify which tasks succeeded or failed, then decide how to handle failures in your result aggregation.

Issue a single message with multiple Task tool calls to trigger single-message coordination. This approach runs all tasks concurrently, balances workload across available agents, and applies an upper bound on parallelization to prevent resource exhaustion.