# Coordinator-subagent orchestration

*Claude Certified Architect - Foundations Study Guide · Agentic Architecture & Orchestration · Last updated October 22, 2026 · https://www.certifyforce.com/study-guide/claude-certified-architect-foundations/agentic-architecture-and-orchestration/coordinator-subagent-orchestration*

How a coordinator agent routes, scopes and checks the work of specialist subagents: hub-and-spoke communication, description-based selection, effort scaling, scope partitioning, the gap-filling refinement loop, when multi-agent is worth its token cost, and how to cap and trace the agent tree.

- Certification: Claude Certified Architect - Foundations
- Exam section: Agentic Architecture & Orchestration
- Study guide: https://www.certifyforce.com/study-guide/claude-certified-architect-foundations

**Official exam objective:**

> Orchestrate multi-agent systems with coordinator-subagent patterns

## What the exam expects

- Design a coordinator that owns decomposition, delegation, routing, error handling, logging and aggregation, with all subagent traffic flowing through it.
- Make the coordinator select only the subagents a query needs and scale effort to query complexity, triggered by clear `description` text or by naming the subagent.
- Partition scope across subagents to avoid duplicated work, and recognise over-narrow decomposition as the cause of missing coverage.
- Add an iterative refinement loop: evaluate the synthesis for gaps, re-delegate targeted queries, re-synthesize.
- Decide when multi-agent is worth its token cost, and cap and trace the agent tree with enforced limits.

Key terms: **coordinator** (lead agent that plans, delegates and synthesizes); **hub-and-spoke** (subagents talk only to the coordinator, never to each other); **effort scaling** (subagent count and tool calls matched to query complexity); **spawn depth** (how many layers of subagents may nest).

## Key facts

| Fact | Detail | Source |
|---|---|---|
| What triggers delegation | Claude matches the task to each subagent's `description` | [Subagents in the SDK](https://code.claude.com/docs/en/agent-sdk/subagents) |
| Guaranteeing a subagent | Name it in the prompt; bypasses automatic matching | [Subagents in the SDK](https://code.claude.com/docs/en/agent-sdk/subagents) |
| Effort scaling | 1 agent / 3-10 calls; 2-4 subagents / 10-15 calls each; 10+ subagents for complex research | [Multi-agent research system](https://www.anthropic.com/research/multi-agent-research-system) |
| Delegation contents | Objective, output format, tools and sources, task boundaries | [Multi-agent research system](https://www.anthropic.com/research/multi-agent-research-system) |
| Long-run coordinator state | Save plan, constraints and finished subtasks to external memory; re-read after truncation | [Multi-agent research system](https://www.anthropic.com/research/multi-agent-research-system) |
| Token cost | Agents about 4× chat; multi-agent about 15× chat | [Multi-agent research system](https://www.anthropic.com/research/multi-agent-research-system) |
| Spawn depth default | 3 layers; `1` stops nested spawning | [Subagents in the SDK](https://code.claude.com/docs/en/agent-sdk/subagents) |
| Concurrent subagents default | 20 | [Subagents in the SDK](https://code.claude.com/docs/en/agent-sdk/subagents) |
| Spend cap | `max_budget_usd` includes subagents; ends with `error_max_budget_usd` | [Subagents in the SDK](https://code.claude.com/docs/en/agent-sdk/subagents) |
| Spawn detection | `tool_use` name `"Agent"` (match `"Task"` too) | [Subagents in the SDK](https://code.claude.com/docs/en/agent-sdk/subagents) |
| Message attribution | `parent_tool_use_id` on messages from inside a subagent | [Subagents in the SDK](https://code.claude.com/docs/en/agent-sdk/subagents) |

## Check yourself

- Why route every subagent message through the coordinator?
- Why should a coordinator in a multi-hour run persist its plan and the user's constraints outside its context?
- When should a prompt name a subagent explicitly instead of relying on its description?
- When is a multi-agent design a poor fit despite its power?
- Which field lets a tracing dashboard attribute streamed messages to the subagent that produced them?

## Sources

- [How we built our multi-agent research system](https://www.anthropic.com/research/multi-agent-research-system)
- [Subagents in the SDK](https://code.claude.com/docs/en/agent-sdk/subagents)
- [Building effective AI agents](https://www.anthropic.com/research/building-effective-agents)
- [Configure permissions](https://code.claude.com/docs/en/agent-sdk/permissions)
- [Claude Certified Architect – Foundations Exam Guide v1.0](https://everpath-course-content.s3-accelerate.amazonaws.com/instructor%2F6nizmqk8tpzpfjvt6qmmav7rh%2Fpublic%2F1783542750%2FClaude+Certified+Architect+%E2%80%93+Foundations+Exam+Guide.pdf)

## About the full unit

The "Core concepts", "Exam traps" and "Worked scenario" sections of this unit, its 14 linked practice questions, and unlimited timed mock exams are included with the Claude Certified Architect - Foundations pack (USD $19): https://www.certifyforce.com/practice-exams