dispatching-parallel-agents

Running parallel agents for concurrent investigation involves dispatching one subagent per independent domain to investigate and fix unrelated failures simultaneously. This approach applies specifically when 2+ failures exist in separate subsystems with no shared state or ordering dependencies.

Concurrent investigation is appropriate for debugging workflows when you have 2+ unrelated failures across different subsystems that have no shared state or ordering dependencies, such as separate failing test files. It enables faster resolution by parallelizing the investigation and fixes.

Identifying independent problem domains requires checking that failures have no shared state or ordering dependencies between them. You must verify that areas like separate failing test files or subsystems are completely unrelated before assigning them to dedicated agents for parallel execution.

Integrating results after parallel subagent dispatch involves a review workflow that consolidates agent summaries and applies non-conflicting fixes in a coordinated way. The standardized process ensures that concurrent investigation outputs are reviewed before final integration.

Parallel dispatch is not suitable for failures with shared state or ordering dependencies. The approach requires completely independent domains to function correctly. Applying it to related failures risks conflicts during the review and integration of concurrent agent fixes.

The best way to manage multiple agents fixing separate subsystem failures is dispatching one subagent per independent domain using a standardized agent prompt. This agent management strategy ensures focused tasks are executed concurrently and results are reviewed systematically.