Dispatching Parallel Agents

Dispatching Parallel Agents

When to Use

• 3+ independent test file failures.
• Multiple broken subsystems (e.g., RAG vs. UI vs. API).
• No shared state or data dependencies between the tasks.

Prerequisites

• Multiple independent failures identified.
• Context is clear for each failure.

Process

The process involves identifying independent failures and dispatching specialists to handle them in parallel.

The Pattern

1. Segment: Group failures by problem domain (e.g., "RAG Pathing", "Plane Auth").
2. Tasking: Create focused, self-contained prompts for each domain.
3. Dispatch: Run agents concurrently using the run or Task primitive.
4. Integrate:
   • Review summaries from all agents.
   • Check for file conflicts (unlikely if segmented properly).
   • Run the FULL test suite to confirm collective success.

Best Practices

Adhere to these best practices to avoid race conditions and context pollution.

Red Flags

• Dependencies between failures: Fix one, might fix the other? Don't parallelize.
• Exploratory debugging: Don't know what's broken yet? Investigate first.
• Shared state: Editing the same files? Sequential only.