Introduction to Sub-agents Concept

What Are Sub-agents?

Sub-agents are specialized Claude instances that a main agent can create and delegate tasks to. Rather than one Claude instance handling every aspect of a complex workflow, the main agent acts as an orchestrator — breaking the task into parts and assigning each part to a sub-agent with the right focus and tools.

Why Sub-agents?

Two fundamental constraints make sub-agents valuable:

• Context window limits: Every Claude instance has a finite context window. For very large tasks, the context fills up. Sub-agents have their own fresh context windows, allowing parallel work on different parts of the problem.
• Specialization: Different parts of a task benefit from different instructions, tools, and focus. A sub-agent tasked with security review can be given security-specific tools and instructions without those cluttering the context of the sub-agent doing documentation.

The Orchestrator Pattern

The main agent acts as a project manager: it plans the overall approach, creates sub-agents with specific mandates, collects their outputs, synthesizes the results, and handles coordination. Sub-agents are focused workers: they receive a specific task, complete it, and return results to the orchestrator.

When to Use Sub-agents

Sub-agents are appropriate when:

• Tasks can be meaningfully parallelized (independent work that doesn't need to coordinate in real time)
• Different parts of the task require different specialized context or tools
• The overall task is too large to fit in a single context window
• You want to isolate risky operations in a contained sub-agent context

For simple, sequential tasks, sub-agents add overhead without benefit. The overhead of creating and coordinating sub-agents is only worthwhile when the task genuinely warrants it.