What Are Subagents

A subagent is a helper agent that the main AI agent delegates a specific piece of work to. Much as a conductor assigns parts to individual performers, the main agent hands research, review, or transformation work to a specialist.

Why Subagents Help

When one agent handles research, implementation, and review together, the conversation history grows and the original instruction fades by the later stages. Subagents address this by isolating context per task.

Scroll horizontally→

Aspect	One agent does everything	Delegated to a subagent
Context	History from every step accumulates	Only the information needed for that task is passed
Clarity of instruction	Multiple roles blur the criteria	The instruction targets a single role
Handling of results	Every intermediate step remains	Only the conclusion returns to the main agent
Review	The agent evaluates its own work	An independent perspective evaluates it

This table scrolls horizontally. Keyboard users can focus the table and use the left and right arrow keys.

The Basic Delegation Shape

graph LR
    Main["Main agent\nmanages overall progress"]
    Sub1["Subagent A\nresearch"]
    Sub2["Subagent B\nreview"]

    Main -->|instruction and input| Sub1
    Sub1 -->|conclusion only| Main
    Main -->|instruction and input| Sub2
    Sub2 -->|findings only| Main

What goes to a subagent is the instruction, input, and completion condition for that task. Passing the entire main conversation history defeats the purpose of isolating context.

Good Fits and Poor Fits

Good fits

• Research that covers a wide area and returns only the conclusion
• Review that checks a deliverable from an independent perspective
• Parallel work that repeats the same processing across different targets

Poor fits

• Simple processing that finishes in one or two steps, where delegation costs more than it saves
• Work that requires continuously sharing fine-grained context with the main agent

Pages in This Section

Scroll horizontally→

Page	Content
Delegation Patterns	What to split out and what to pass along
Context Isolation	Why isolation improves quality, and its caveats
When Not to Use Subagents	Cases where delegation becomes inefficient

This table scrolls horizontally. Keyboard users can focus the table and use the left and right arrow keys.

Summary

• A subagent is a helper agent that the main agent delegates specific work to
• The main benefit is isolating context per task
• Pass only the instruction, input, and completion condition, and receive the conclusion
• Simple work and work needing shared context are poor fits

Related Links

• What Is Orchestration
• What Is an AI Agent