Originally published at claudeguide.io/claude-multi-agent-orchestration

Claude Multi-Agent Orchestration: Patterns for Complex Workflows

Multi-agent systems use multiple Claude instances working together — an orchestrator agent that plans and delegates, and subagent specialists that execute specific tasks. The key design principle: subagents should be narrow and reliable; orchestrators should be broad and strategic in 2026. A web research agent, a data analysis agent, and a report-writing agent each do one thing well. An orchestrator decides which to use, in what order, and how to combine their outputs. This guide covers the core patterns.

---

When to use multi-agent vs single-agent

Single agent (simpler, default): when the task fits in one context window, doesn't benefit from specialisation, and doesn't need parallel execution.

Multi-agent (more complex, when justified):

• Task is too long for one context window
• Sub-tasks benefit from specialised agents with distinct system prompts
• Sub-tasks can execute in parallel (faster completion)
• Reliability improves when agents cross-check each other's work

Don't use multi-agent systems for simple tasks — the orchestration overhead adds latency and cost.

---

Pattern 1: Sequential pipeline

Tasks depend on each other; execute in order:

python
import anthropic
from dataclasses import dataclass
from typing import Callable

client = anthropic.Anthropic()

@dataclass
class AgentConfig:
    name: str
    system_prompt: str
    model: str = "claude-sonnet-4-5"
    max_tokens: int = 4096

def run_agent(config: AgentConfig, input_text: str) -

[→ Get the Agent SDK Cookbook — $49](https://shoutfirst.gumroad.com/l/ogxhmy?utm_source=claudeguide&utm_medium=article&utm_campaign=claude-multi-agent-orchestration)

*30-day money-back guarantee. Instant download.*