Recipe Overview

Complex tasks with unpredictable subtasks require dynamic breakdown. An orchestrator-workers agent acts as a manager LLM that decomposes the input on the fly and assigns subtasks to other agents. Anthropic calls this ideal when you can't predict the subtasks needed. For example, in a large coding problem, the orchestrator may first decide to write a user interface module and then backend logic, each handled by specialized worker agents. This solves coordination by having a central planner that can adapt the workflow based on intermediate results, making it suitable for open-ended or exploratory tasks.