Production reference for dev teams shipping agents

25+ production-testedmulti-agent patterns.Full code. Two SDKs.

The Agent Orchestration Cookbook is a production reference for dev teams shipping multi-agent AI systems — research agents, code-review agents, data-pipeline agents, customer-support triage, and the core orchestration patterns underneath. Each pattern ships with full code in Claude Agent SDK and LangGraph, in both Python and TypeScript, plus evaluation harnesses, failure-mode notes, and cost-per-run analysis.

Get the Cookbook — $79one-time · full source, run it in your repo · 30-day refund
25+ patterns · two SDKs
25+ Orchestration Patterns
Production
Full Code · SDK + LangGraph
TS · PY
Eval Harness per Pattern
Tested
Failure-mode + Cost Notes
Per run
Runs in
Python · TypeScript · Claude Agent SDK · LangGraph
01.The Problem

Every team is building agents. Almost everyone is reinventing the same patterns badly.

Your team ships its first agent. It works in the demo. It fails in three weeks of production because nobody handled the retry-with-reflection edge case where the model loops forever. You patch it. The next agent ships. Same loop happens.

Multi-agent patterns are not new. The orchestrator-worker pattern, the hierarchical-supervisor pattern, the parallel map-reduce — these are solved problems with known failure modes. The problem is the knowledge lives in Anthropic engineering blogs, LangChain Discord threads, a handful of conference talks, and the memories of senior engineers who learned the hard way.

The cookbook is what your senior engineer would build on their best week — written down, tested, with cost numbers and failure notes. Drop the patterns into your repo. Stop reinventing.

The reinvention pattern · Symptoms
  • 01Engineer writes a multi-step agent in 4 days; works in dev
  • 02Production: agent loops forever on edge case in ticket 23
  • 03Team adds max_iterations=10 as a band-aid; ticket 47 hits limit silently
  • 04Adds reflection step that doubles token costs without measurable quality gain
  • 05Reads a blog post about supervisor patterns; rewrites half of it
  • 06Six weeks later: still no eval harness, no failure-mode docs, no cost dashboard
02.What's Inside

Six deliverables. Engineered to ship together.

01

The 25+ Pattern Library

Organized into five categories: core orchestration, research agents, code & engineering agents, data pipeline agents, customer operations agents. Each pattern is self-contained and composable.

02

Full Implementation Code

Each pattern in four paths: Claude Agent SDK (Python + TypeScript), LangGraph (Python + TypeScript). Drop-in ready. Real implementations, not pseudocode or architecture diagrams.

03

Evaluation Harnesses

Per-pattern eval scripts with golden datasets, scoring criteria, regression test scaffolds. Includes provider-agnostic benchmarks for Claude, GPT, and Gemini comparisons.

04

Failure-Mode Notes

For each pattern: where it breaks in production, why, what to do about it. The notes you'd otherwise pay senior engineers to learn through outages.

05

Cost-Per-Run Analysis

Token usage, API costs across providers, latency profiles (p50, p95), retry-rate impact. Pick patterns that match your unit economics, not just your accuracy targets.

06

Architecture Decision Records

Why graph vs. hierarchical. When subagents beat tools. When to use Claude SDK vs. LangGraph. The senior-engineer reasoning behind every pattern, written down so you can defend the choice in code review.

03.The 25+ Patterns

Five categories. Twenty-five patterns. Every one in four implementation paths.

01

Core Orchestration Patterns

8 patterns
  • ·Sequential pipeline
  • ·Orchestrator-worker
  • ·Hierarchical supervisor
  • ·Parallel map-reduce
  • ·Handoff routing
  • ·Voting / ensemble
  • ·ReAct with memory
  • ·Recursive decomposition
02

Research & Information Agents

4 patterns
  • ·Web research agent
  • ·Multi-source synthesis
  • ·Citation-tracked research
  • ·Iterative literature review
03

Code & Engineering Agents

4 patterns
  • ·Code review agent
  • ·Refactor planner
  • ·Test generation
  • ·Bug triage with root-cause
04

Data Pipeline Agents

4 patterns
  • ·ETL coordinator
  • ·Data validation agent
  • ·Schema inference
  • ·Anomaly detection
05

Customer Operations Agents

5 patterns
  • ·Support ticket triage
  • ·Intent classifier + router
  • ·Escalation logic
  • ·FAQ generation
  • ·Sentiment-aware response
+1

Pattern Composition Guide

bonus

Most production agent systems combine 3 to 5 patterns. The composition guide shows common combinations and the interface gotchas that emerge when patterns meet.

Pattern count is 25 across the five categories plus the composition guide. New patterns are added when material new orchestration approaches emerge (and survive 90 days in production somewhere). Existing buyers get those additions included in the lifetime updates.

04.Inside a Pattern

Sample: the Customer Support Triage pattern. Code, evals, failure notes, cost.

pattern · customer_support_triage
orchestrator-worker

Customer Support Triage

A classifier agent routes incoming tickets to specialist agents based on detected intent. Specialist agents handle their domain end-to-end, with optional subagent delegation for complex cases. The classifier is cheap (Haiku); the specialists are capable (Sonnet). Cost-efficient and accurate at scale.

Incoming ticket
   │
   ▼
┌──────────────┐
│  Classifier  │ ← Haiku 4.5
│  (intent)    │
└──────┬───────┘
       │
       ├──→ Billing agent      ← Sonnet 4.6
       ├──→ Technical agent    ← Sonnet 4.6 + subagents
       └──→ Account agent      ← Sonnet 4.6
claude-sdk · python · customer_support_triage.py
excerpt
from claude_agent_sdk import Agent
from typing import Literal

Intent = Literal["billing", "technical", "account", "unknown"]

classifier = Agent(
    model="claude-haiku-4-5",
    system="Classify the ticket. Return one of: billing, technical, account.",
    output_schema={"intent": Intent, "confidence": float},
)

specialists = {
    "billing":   Agent(model="claude-sonnet-4-6", system=BILLING_SYS_PROMPT),
    "technical": Agent(model="claude-sonnet-4-6", system=TECHNICAL_SYS_PROMPT,
                       subagents=[diagnostics_subagent]),
    "account":   Agent(model="claude-sonnet-4-6", system=ACCOUNT_SYS_PROMPT),
}

async def triage(ticket: str) -> str:
    result = await classifier.run(ticket)
    if result.intent == "unknown" or result.confidence < 0.7:
        return await human_handoff(ticket, reason="low_confidence")
    return await specialists[result.intent].run(ticket, context=result)
eval-harness
CI · #847
$ pattern-eval customer_support_triage

PATTERN     customer_support_triage
DATASET     golden/support_v2.jsonl (n=412)
─────────────────────────────────────
Accuracy           92.7%
Routing precision  94.1%
Cost / ticket      $0.043
p50 latency        1.8s
p95 latency        2.4s
Escalation rate    7.3%
─────────────────────────────────────
exit: 0
Failure modes
  • ·Short tickets (<3 words). Classifier hits low-confidence threshold; routes 18% to human handoff.
  • ·Multi-issue tickets. Classifier picks dominant intent; secondary issue lost. Mitigation in §3.2.
  • ·Specialist hallucination. Technical agent makes up account details. Mitigated by RAG + structured output.
cost-per-run · per ticket
directional
ComponentModelCost / ticketNotes
ClassifierHaiku 4.5$0.001Short input, structured output
Billing specialistSonnet 4.6$0.038~38% of tickets
Technical specialistSonnet 4.6 + subagent$0.072~22% of tickets
Account specialistSonnet 4.6$0.027~33% of tickets
Human handoff$0~7% of tickets
Weighted avg$0.043per ticket

One pattern. One of 25. Each one ships with the same depth: architecture, code in four implementation paths, eval harness, failure-mode notes, cost-per-run table. Your senior engineer’s best week, written down.

05.Platform Coverage

Two platforms. Two languages. Four implementation paths per pattern.

Two platforms hits roughly 80% of the serious multi-agent market with manageable scope. Claude Agent SDK is the Anthropic-native runtime that powers Claude Code — increasingly the default for production agent work. LangGraph is the dominant open-source orchestration framework and is provider-agnostic, so the LangGraph implementations work with Claude, GPT, or Gemini. Pick the path that matches your stack. Ignore the rest.

Claude Agent SDK
Python · Python 3.10+
$ pip install claude-agent-sdk

Anthropic-committed stack. Deep tool-use, hooks, subagents, MCP. The runtime that powers Claude Code.

Claude Agent SDK
TypeScript · Node.js 18+
$ npm install @anthropic-ai/claude-agent-sdk

Anthropic-committed stack on Node. Feature-parity with the Python SDK. V1 stable interface targeted.

LangGraph
Python · Python 3.10–3.14
$ pip install langgraph

Provider portability. Graph-based state machines. Most production-proven open-source orchestration in market.

LangGraph
TypeScript · Node.js 18+
$ npm install @langchain/langgraph

Provider portability on Node. Feature parity with Python on core capabilities (StateGraph, checkpointing, streaming, human-in-the-loop).

The patterns themselves translate. If your stack is on a different framework — OpenAI Agents SDK, Mastra, CrewAI, bare provider APIs — the architectures, failure-mode notes, and cost analysis still apply. You’ll port the code yourself, but you don’t reinvent the patterns.

06.What's Not in the Cookbook

Be honest about what the $79 buys you. And what it doesn’t.

What’s in
  • 25+ patterns, full code in four implementation paths.
  • Evaluation harnesses with golden datasets and scoring criteria.
  • Failure-mode notes from real production deployments.
  • Cost-per-run analysis with token, latency, and retry-rate breakdowns.
  • Architecture decision records explaining why each pattern was designed this way.
What’s out
  • Not a managed platform. No UI, no hosted runtime, no observability dashboard. The cookbook is code and documentation; deployment is your problem.
  • Not OpenAI Agents SDK, Mastra, or CrewAI. Two platforms deliberately. Patterns translate; ports are your problem.
  • Not voice agents. Different product category. See the Voice Agent Deployment Kit if that’s your scope.
  • Not fine-tuning or training. The cookbook assumes you’re using off-the-shelf models. Custom training is a separate discipline.
  • Not a RAG cookbook. Some patterns use retrieval, but RAG architecture is a different category with its own cookbook-shaped book somewhere out there.

The cookbook is one thing — a pattern reference for multi-agent orchestration — and it does that one thing thoroughly. The boundaries are deliberate. Other RedHub AI products cover other parts of the AI engineering stack.

07.Who This Is For

Be honest with yourself before you click buy.

Buy if
  • Your team ships multi-agent features in production or is about to.
  • You’re using Claude Agent SDK, LangGraph, or considering either.
  • You’ve already reinvented at least one pattern badly.
  • You’re a senior engineer, AI/ML engineer, eng manager, or technical founder.
  • You want production-tested patterns to own, not blog-post snippets to translate.
Skip if
  • You’re using OpenAI Agents SDK exclusively (patterns translate, code doesn’t).
  • You haven’t shipped a single agent to production yet — start simpler.
  • You want a managed platform with a UI and dashboards.
  • You want done-for-you implementation, not a cookbook to own.
  • Nobody on your team reads TypeScript or Python.
08.Pairs Well With

Orchestrate the agents, then prove they hold.

The patterns here stitch agents together; the rest of the reliability stack proves they work. The Agent Reliability Harness stress-tests the orchestrated flow end-to-end, the Prompt Eval System scores the prompts each step depends on, and the RAG Retrieval Grader grades the retrieval feeding them.

09.Common Questions

The questions engineers actually ask before pulling out the company card.

Two platforms hits roughly 80 percent of the serious multi-agent market with manageable scope. Claude Agent SDK is the Anthropic-native runtime that powers Claude Code — increasingly the default for production agent work. LangGraph is the dominant open-source orchestration framework and is provider-agnostic, so the LangGraph implementations work with any model (Claude, GPT, Gemini). OpenAI Agents SDK is excellent but OpenAI-locked. Mastra is the strongest TypeScript-native framework but newer. CrewAI is great for role-based prototyping but lighter on production controls. The patterns themselves translate — if you're on a different stack, the architectures and failure-mode notes still apply.

Instant access · Lifetime updates · 30-day refund

Stop reinventing patterns.
Ship the ones that work.

$79 once. 25+ patterns. Full code in Claude Agent SDK and LangGraph, Python and TypeScript. Eval harnesses, failure notes, cost analysis. Your senior engineer’s best week, written down.

Sold by RedHub AI LLC · Secured by Stripe · redhub.ai