Coordinator-Subagent Patterns

Hub-and-spoke architecture where a coordinator agent manages all inter-subagent communication, decomposes tasks, delegates to specialized subagents, and synthesizes results. Understand why subagents have isolated context and how the coordinator prevents coverage gaps.

Quick Reference

→Hub-and-spoke: coordinator is the central hub, subagents are spokes -- no direct spoke-to-spoke communication
→Subagents have isolated context -- they do NOT inherit the coordinator's conversation history
→The coordinator's job: decompose the task, delegate to subagents, aggregate and synthesize results
→Overly narrow decomposition causes coverage gaps -- the 'creative industries' trap
→Iterative refinement: coordinator evaluates synthesis, identifies gaps, re-delegates as needed
→Each subagent gets a focused system prompt and restricted tool set appropriate to its subtask
→Subagent results flow back ONLY to the coordinator, which decides what to share with other subagents
→Partition scope to minimize overlap -- overlapping scopes lead to contradictory findings
→The coordinator must validate coverage before producing the final response
→Use parallel subagent invocation when subtasks are independent

Hub-and-Spoke Architecture

Definition

In the hub-and-spoke pattern, a single coordinator agent (the hub) manages all communication with subagents (the spokes). Subagents never communicate directly with each other. The coordinator decomposes the task, delegates subtasks, collects results, and synthesizes the final output.

Hub-and-spoke: a coordinator delegates to specialized subagents

The hub-and-spoke pattern is the primary multi-agent architecture tested on the exam. Its defining characteristic is centralized control: the coordinator is the only entity that sees the full picture. Each subagent operates in isolation, focused on its specific subtask, unaware of what other subagents are doing. This isolation is a feature, not a bug -- it prevents subagents from being distracted by irrelevant context and allows the coordinator to maintain a coherent overall strategy.

Think of it like a research team lead managing specialists. The lead assigns each researcher a specific topic, collects their findings, identifies gaps or contradictions, and writes the final synthesis. The researchers do not coordinate directly -- the lead ensures coherence.

Coordinator vs Subagent Responsibilities

Responsibility	Coordinator	Subagent
Task decomposition	Breaks the user's request into focused subtasks	Receives a single, well-defined subtask
Context management	Maintains the full picture; decides what context each subagent needs	Operates only with the context explicitly provided
Tool access	Has access to the Task tool for spawning subagents	Has access to domain-specific tools (search, database, API, etc.)
Communication	Routes information between subagents as needed	Reports results back to coordinator only
Quality control	Evaluates completeness, identifies gaps, re-delegates	Focuses on depth and accuracy within its scope
Synthesis	Combines subagent results into coherent final output	Produces focused findings for its subtask
Scope management	Ensures full coverage without overlap	Stays within assigned scope boundaries

Context Isolation: Why It Matters

One of the most important concepts tested on the exam is that subagents have isolated context. When the coordinator spawns a subagent, that subagent does NOT automatically inherit the coordinator's conversation history, system prompt, or prior tool results. The subagent starts with a clean slate and only knows what the coordinator explicitly passes to it.

Sign in to read this article

This is a premium article. Sign in with your Google account to continue.