ultrawork

Parallel execution dispatches independent subtasks simultaneously to multiple agents at appropriate model tiers. It applies when you have several independent subtasks that can run concurrently, allowing you to manage completion yourself while optimizing cost and performance through tier selection.

Choosing the right agent tier involves reading the agent-tier guidance before delegation to optimize cost and performance. You route each independent subtask to the LOW, STANDARD, or THOROUGH tier based on its complexity and required execution depth.

The best way to execute long-running operations without blocking foreground tasks is using run_in_background during delegation. This dispatches lengthy operations to background agents while quick tasks proceed in the foreground, maximizing concurrency and throughput.

Parallel task execution does not handle state management or persistence. It strictly provides high-throughput execution by dispatching independent subtasks to multiple agents, leaving persistence and state management entirely to the user who manages task completion themselves.

You should not use concurrent agent delegation when subtasks are dependent on each other. This parallel execution approach applies only when you have several independent subtasks that can run concurrently without interaction, as it lacks built-in state management for coordinating dependencies.