dispatching-parallel-agents

To execute parallel tasks concurrently, dispatch one agent per independent problem domain. This allows each task to be tackled in isolation without shared state, eliminating the wasted time of sequential investigation.

Use parallel agents when you face 3 or more independent failures or tasks that can be resolved in isolation. It enforces per-agent task tracking, defined scopes for each agent, and parallel execution with a consolidated post-run review.

Tracking progress during parallel execution is handled by centralized task tracking. This mechanism captures who is working on what and their current status, ensuring progress visibility across all dispatched agents.

Parallel agents are designed for tasks without shared state. You should not use this approach for dependent failures, as each agent operates within an isolated, well-defined boundary to avoid shared state.

The best way to handle multiple unrelated failures is dispatching parallel agents. It assigns a dedicated agent to each problem, enforces defined scopes, and provides a consolidated post-run review of all independent results.

Dispatching parallel agents applies to 3 or more independent failures or tasks. For orchestrating just two concurrent problems, a simpler multi-tasking approach without centralized task tracking may be more appropriate.