agent-orchestration

To run parallel agent workstreams, you coordinate multiple tasks using a decision framework that isolates contexts and manages background sub-agents. This approach maximizes throughput by reducing idle time and splitting independent tasks for concurrent execution.

The best way to split work for parallel execution is using worktree setup to create isolated working directories. This pattern identifies tasks that can safely run concurrently versus those requiring serial processing, then manages sub-agents within those isolated branches.

Use parallel agent coordination when independent tasks can safely run concurrently to maximize throughput. A decision framework evaluates your tasks to identify what can execute in parallel versus what must remain serial, ensuring safe and efficient workstream coordination.

You can execute parallel agents within a single session or across multiple terminals. The coordination strategy applies patterns for both approaches, managing background tasks and sub-agents with clear instructions, guardrails, and safe cleanup.

To safely clean up sub-agents after parallel workstreams, follow the guardrails and cleanup instructions built into the coordination framework. These ensure background tasks are properly terminated and isolated worktree contexts are synthesized or removed after execution.