dispatching-parallel-agents

Run parallel investigations by dispatching one subagent per independent domain to investigate failures concurrently. Ensure each agent has clearly scoped, non-overlapping tasks and no shared state or ordering dependencies before execution.

Dispatch parallel agents when you have 2+ unrelated failures, such as separate failing test files or subsystems. Requirements include clearly scoped agent tasks, non-overlapping changes, and no shared state or ordering dependencies between the failures.

You can use parallel agents to fix independent bugs across different subsystems by distributing work across multiple agents. The orchestration enforces domain scoping rules to avoid cross-team interference and merge conflicts during concurrent execution.

The best way to orchestrate agents for concurrent investigation is to dispatch one subagent per independent domain, enforce non-overlapping changes, and apply a structured review and integration flow to consolidate results and ensure green deployments.

Limitations of dispatching parallel agents include requiring 2+ unrelated failures with no shared state or ordering dependencies. Agents cannot handle overlapping changes or cross-domain state, and consolidated integration checks are required after parallel execution completes.