dispatching-parallel-agents

To investigate independent failures in parallel, dispatch one agent per problem domain so multiple investigations happen concurrently. This pattern requires clear scoping per agent and non-shared state to reduce total time to resolution.

Dispatch parallel agents for troubleshooting when you have 3 or more unrelated failures or separate subsystems. This approach is effective when these independent problems can be analyzed without shared context between the dispatched agents.

The best way to structure parallel agent tasks is to identify independent problem domains, create focused tasks for each, and dispatch them concurrently. Agents must produce evaluative outputs that are easy to merge into a final centralized report.

Dispatching parallel agents does not interfere with production code. The pattern explicitly requires non-interference with production code, ensuring each focused investigation operates independently without altering shared state.

You cannot use parallel agents to analyze subsystems that share context. This approach strictly requires independent problem domains with non-shared state, meaning investigations must happen without shared context to ensure accurate, mergeable summaries.

Agent outputs should be concise, integration-ready summaries formatted for centralized review. These evaluative outputs must be easy to merge into a final report after the parallel investigations conclude.