Skip to content

Definition

A workflow where a central orchestrator LLM dynamically breaks down tasks, delegates them to worker LLMs, and synthesizes their results.

Key points

  • Key difference from Parallelization Workflow: subtasks are not pre-defined — the orchestrator determines them based on the input.
  • When to use: "complex tasks where you can't predict the subtasks needed" (e.g., number/nature of file changes in a coding task).
  • Maps directly onto Hub-and-spoke orchestration — orchestrator = hub, workers = spokes.

Why it matters for the exam

  • "Orchestrator-workers" is a named workflow pattern; the testable distinction is dynamic, runtime-determined subtasks vs the fixed subtasks of parallelization.

Common gotchas

  • Start simple; only add agentic complexity when it demonstrably improves outcomes — use it only when you genuinely cannot predict the subtasks.

See also

Sources