ultrawork

To run multiple agents in parallel, invoke them simultaneously with explicit model selection and prompts. This parallel execution targets high-throughput task orchestration, enabling multiple independent tasks to complete with minimal manual coordination.

Smart model routing in Gemini-based workflows uses tiered agents to save tokens. It requires reading the agent tier matrix to select the appropriate model before delegating work, ensuring token-efficient execution across parallel tasks.

Yes, you can use background execution for long-running tasks by setting run_in_background to true. This allows long operations to execute without blocking other agents in the parallel orchestration pipeline.

Yes, explicit model selection is a critical requirement when orchestrating agents. You must always pass the model parameter explicitly to ensure the correct agent tier is used for token-efficient routing in your workflow.

Parallel task orchestration without persistence lacks state management, verification loops, and don't-stop-until-done behavior. You must use complementary tools like ralph or autopilot to add session persistence and mandatory architect verification.

You should not use parallel-only execution when your workflow requires state management, mandatory architect verification, or persistent session handling. In these cases, use ralph for persistent parallel work or autopilot for autonomous parallel work instead.