Session 5: AI Agents, Workflows & n8n

Date: October 17, 2025
Duration: 2 hours
Participants: WiseTech Global Team
Recording: Watch Session Recording

📋 Overview

In this comprehensive session, we explored the landscape of AI application development, from understanding where startup dollars go in AI spending to building practical AI agents with n8n. The session covered context engineering principles, AI agent fundamentals, and provided hands-on demonstrations of creating intelligent assistants with memory and tool capabilities using the Amap (高德地图) MCP service.

🎯 Key Topics Covered

1. AI Application Spending Report

Understanding the economics of AI applications is crucial for strategic decision-making.

Key Reports:

Key Insights:

Infrastructure costs dominate early-stage AI applications
Token consumption patterns reveal usage optimization opportunities
Strategic resource allocation is critical for AI product success

2. Context Engineering

Context engineering is the art and science of optimizing how we provide information to AI models for better results.

Featured Resources:

Core Principles:

Precision over Volume: More context isn't always better - relevance matters
Structured Information: Organize context hierarchically for better comprehension
Dynamic Context: Adapt context based on conversation flow and user intent
Memory Integration: Leverage conversation history effectively

3. AI Agents & Workflow Platforms

AI Agents Ecosystem

Platform Comparison:

Platform	Type	Strengths	Use Cases
ChatGPT AgentKit	Framework	Official OpenAI support	Prototyping, research
LlamaIndex	Framework	Data integration focus	RAG applications
Firecrawl	Tool	Web scraping + agents	Data collection agents
n8n	Workflow	Visual, open-source	Business automation
Coze	Platform	Low-code	Rapid prototyping
Dify	Platform	Enterprise features	Production deployments

Key Resources:

4. Understanding n8n

From "nodemation" to "n8n"

The founder Jan Oberhauser originally wanted to name this tool nodemation.

Name Construction:

markdown

nodemation = node + automation
    │          │        │
    │          │        └─ Automation
    │          └────────── Node (dual meaning)
    │
    └─ ① Built with Node.js
       ② Node-based visual interface

But Jan thought the name was too long and not concise enough. So he used a clever approach:

markdown

nodemation
n  o  d  e  m  a  t  i  o  n
└───────── 8 letters ─────────┘

↓ Replace middle 8 letters with the number 8

n8n ✨

Numeronym Examples:

Original	Abbreviation	Explanation
Kubernetes	K8s	8 letters between K and s
Internationalization	i18n	18 letters between i and n
Localization	l10n	10 letters between l and n
nodemation	n8n	8 letters between n and n

5. Traditional AI vs AI Agent

Traditional AI:

You ask: "What's the weather today?"
It answers: Weather information

AI Agent:

You say: "Help me plan a weekend trip"
It will:
- Check weather forecast
- Recommend suitable destinations
- Find transportation options
- Even book hotels for you

Core Difference:

Traditional AI = Information retriever
AI Agent = Autonomous thinking and acting intelligent assistant

Formula:

markdown

Agent = Loop (LLM + Tools + Context + Memory)

Capability Comparison:

Dimension	Traditional AI	AI Agent
Response Mode	Passive answers	Active execution
Processing Depth	Single-turn dialogue	Multi-step planning
Decision Making	Fixed process	Dynamic decisions
Capability Range	Single function	Tool invocation

6. Building Your First AI Agent with n8n

Step 1: Understanding Agent Architecture

Three-Layer Architecture:

Trigger Layer: How to start? (Chat/Webhook/Schedule)
Processing Layer: How to think? (LLM + Memory)
Execution Layer: How to act? (Tools)

Modular Design Benefits:

Flexibly replace any layer without affecting others
Quickly locate problems at specific layers
Reuse existing modules to build new systems

Step 2: Adding AI Agent Node

Open n8n workflow canvas
Click "Add Node" button (+)
Search for "AI Agent"
Select the AI Agent node

The AI Agent node comes with a chat trigger, enabling interactive conversation.

Step 3: Three Connection Points

Left: Chat Model - Agent's "brain" (determines intelligence level)
Middle: Memory - Agent's "memory bank" (remembers conversation history)
Right: Tools - Agent's "toolbox" (grants execution capabilities)

7. Adding Memory to Your Agent

Memory Storage Options:

Storage Type	Use Case	Advantages	Disadvantages
Simple Memory	Testing/Development	Zero configuration	Lost on restart
Redis	High concurrency	Fast performance	Requires maintenance
MongoDB	Complex data	High flexibility	Complex setup
PostgreSQL	Enterprise apps	Strong reliability	Resource intensive

Progressive Upgrade Principle:

Start with the simplest (Simple Memory)
Upgrade when encountering limitations
Don't over-design

Memory Configuration:

Context Window Size: 10 (remember last 10 conversation turns)
Session ID: Automatically managed for chat triggers

{{ $json.sessionId }}

8. Integrating Tools - Amap MCP Service

What is MCP?

MCP (Model Context Protocol) is a standardized protocol that allows AI models to safely access external tools and services.

MCP Advantages:

✅ Standardized interface - Unified calling method
✅ Plug-and-play - No complex configuration
✅ Feature-rich - Multiple capabilities in one service

Amap (高德地图) MCP Capabilities

Weather Query - Get real-time weather and forecasts for any city
Address Search - Find specific addresses and coordinates
Nearby Search - Find restaurants, hotels, gas stations, etc.
Route Planning - Calculate optimal travel routes
Administrative Divisions - Get city and district information

9. System Prompt Design

Understanding System Message vs User Message:

System Message (System Prompt):

Defines AI's identity, behavior rules, and working methods
Users can't see it, but it affects all AI responses
Content: Role setting, output format, tool usage rules
Scope: Entire conversation session

User Message (User Prompt):

User's specific questions or instructions
Content: Specific questions, needs, dialogue content
Scope: Current conversation turn

Example System Prompt for Weather Agent:

markdown

You are an intelligent life assistant equipped with Amap tools. Please follow these requirements:

## Basic Principles
1. Always reply in Chinese
2. Maintain professional yet friendly tone
3. Be honest if uncertain about answers

## Tool Usage Guide

### Weather Query
- Keywords: weather, temperature, rain, sunny, etc.
- Action: Call weather query tool
- Output Format: Strictly follow this template for personalized weather forecast

**Weather Forecast Output Template:**
1. 🌤️ Today's Weather Overview
2. 🌡️ Temperature and Feels-Like Temperature
3. 👔 Clothing Suggestions
4. 🚗 Travel Reminders
5. 💡 Life Tips

Requirements:
- Vivid and friendly language suitable for daily reading
- Include practical life advice
- Use emojis for readability
- Total length: 200-300 characters

### Location Query
- Keywords: where, address, how to get, route, nearby, around, find, etc.
- Action: Call address search, nearby search, or route planning tools
- Output Format:
  - Use 📍 to mark address location
  - Use ⭐ for rating information
  - Use 🚶‍♂️🚗 for distance and transportation
  - Provide **specific numbers and distances**
  - Sort recommendations by **distance and rating**

10. Practical Use Cases

Scenario 1: Trip Planning

User: "Going to Shanghai tomorrow for business, check the weather for me"
Agent Actions:
- Calls weather query tool
- Formats response with travel suggestions
- Provides clothing recommendations

Scenario 2: Location Search

User: "Find nearby coffee shops"
Agent Actions:
- Detects current location
- Searches nearby coffee shops
- Returns sorted results with ratings and distances

Scenario 3: Complex Query

User: "I'm at Nanjing South Station, need to go to Confucius Temple, what's the weather like?"
Agent Actions:
- Queries weather for destination
- Calculates route and travel time
- Provides comprehensive travel plan

📊 Presentations & Materials

Session Recording

🎥 Watch Full Session

Recording Highlights:

AI spending report analysis
Context engineering principles
Live n8n agent building demo
Amap MCP integration walkthrough
System prompt design best practices
Real-world use case demonstrations

Demo Configuration

📄 Download n8n Workflow Configuration

This configuration file contains the complete n8n workflow demonstrated in the session, including:

AI Agent node with Amap MCP integration
Simple Memory configuration
System prompt templates
Tool configurations

To import:

Open your n8n instance
Click "Import from File"
Select the downloaded JSON file
Configure your API keys and settings

🔗 Featured Resources

Industry Reports & Research

a16z AI Application Spending Report - Startup spending analysis
OpenAI 1T Tokens Customers - Usage patterns
a16z Report (Chinese) - Translated insights

Context Engineering

YouTube: Context Engineering - Video tutorial
LangChain + Manus (Chinese) - Best practices
Manus 1.5 Release - Latest features

AI Agent Platforms & Tools

ChatGPT AgentKit - OpenAI's official framework
LlamaIndex - Data integration platform
LlamaIndex Multi-Agent - Production framework
Firecrawl - Web scraping tool
Firecrawl Agent Builder - Open source builder
n8n - Workflow automation platform
n8n GitHub - Source code
Coze - Low-code agent platform
Dify - Enterprise AI platform
Dify GitHub - Open source code

🎮 Quiz Activity

Quiz Details

Total Questions: TBD
Duration: 15 minutes
Topics Covered:

AI agent fundamentals
Context engineering principles
n8n workflow design
MCP protocol understanding
System prompt engineering

🏆 Quiz Results & Winners

🥇 First Place

Winner: Leo Qian

🥈 Second Place

Winner: Anthony Liang

🥉 Third Place

Winner: Hunter Yang

📊 Full Results

Rank	Name
🥇 1	Leo Qian
🥈 2	Anthony Liang
🥉 3	Hunter Yang

🔑 Key Insights

Agent Formula: Understanding Agent = Loop (LLM + Tools + Context + Memory) provides a systematic approach to agent design
Modular Architecture: Three-layer architecture (Trigger, Processing, Execution) enables flexible and maintainable agent systems
Progressive Development: Start simple (Simple Memory) and upgrade based on actual needs, avoiding over-engineering
Prompt Engineering: Well-designed system prompts are crucial for consistent and reliable agent behavior
MCP Standardization: Model Context Protocol provides a unified way to integrate external tools and services
Context Engineering: Quality over quantity in context provision leads to better AI responses

💬 Practical Takeaways

For Developers:

Start with n8n's visual interface for rapid prototyping
Use MCP services for standardized tool integration
Apply the three-layer architecture thinking to all agent designs
Test memory configurations progressively

For Product Teams:

Understand AI application spending to optimize costs
Consider agent capabilities when planning AI features
Design user experiences around agent workflows
Leverage existing platforms before building custom solutions

For Everyone:

AI Agents are fundamentally different from chatbots
Tool integration multiplies agent capabilities
Memory enables continuity in conversations
Context engineering improves response quality

📚 Further Learning

Next Steps:

Explore n8n's official documentation and templates
Experiment with different MCP services
Study prompt engineering techniques
Build a personal assistant agent as practice

Recommended Resources:

n8n Academy courses
MCP specification documentation
LangChain agent tutorials
OpenAI prompt engineering guide

🎯 What's Next?

Looking forward to Session 6, where we'll explore:

Advanced agent orchestration patterns
Multi-agent collaboration systems
Production deployment strategies
Monitoring and observability for AI agents

Want to Try This Yourself?

All demo configurations and code samples are available. Start with n8n Community Edition (free) and experiment with building your first agent!

Questions or Feedback?

Reach out to Damien Li or the Crossroad AI team for discussions, clarifications, or collaboration opportunities!

Session 5 | October 17, 2025 | AI Agents, Workflows & n8n | Damien Li

Session 5: AI Agents, Workflows & n8n ​

📋 Overview ​

🎯 Key Topics Covered ​

1. AI Application Spending Report ​

2. Context Engineering ​

3. AI Agents & Workflow Platforms ​

4. Understanding n8n ​

From "nodemation" to "n8n" ​

5. Traditional AI vs AI Agent ​

6. Building Your First AI Agent with n8n ​

Step 1: Understanding Agent Architecture ​

Step 2: Adding AI Agent Node ​

Step 3: Three Connection Points ​

7. Adding Memory to Your Agent ​

8. Integrating Tools - Amap MCP Service ​

What is MCP? ​

Amap (高德地图) MCP Capabilities ​

9. System Prompt Design ​

10. Practical Use Cases ​

📊 Presentations & Materials ​

Session Recording ​

Demo Configuration ​

🔗 Featured Resources ​

Industry Reports & Research ​

Context Engineering ​

AI Agent Platforms & Tools ​

🎮 Quiz Activity ​

Quiz Details ​

🏆 Quiz Results & Winners ​

🥇 First Place ​

🥈 Second Place ​

🥉 Third Place ​

📊 Full Results ​

🔑 Key Insights ​

💬 Practical Takeaways ​

📚 Further Learning ​

🎯 What's Next? ​

Session 5: AI Agents, Workflows & n8n

📋 Overview

🎯 Key Topics Covered

1. AI Application Spending Report

2. Context Engineering

3. AI Agents & Workflow Platforms

4. Understanding n8n

From "nodemation" to "n8n"

5. Traditional AI vs AI Agent

6. Building Your First AI Agent with n8n

Step 1: Understanding Agent Architecture

Step 2: Adding AI Agent Node

Step 3: Three Connection Points

7. Adding Memory to Your Agent

8. Integrating Tools - Amap MCP Service

What is MCP?

Amap (高德地图) MCP Capabilities

9. System Prompt Design

10. Practical Use Cases

📊 Presentations & Materials

Session Recording

Demo Configuration

🔗 Featured Resources

Industry Reports & Research

Context Engineering

AI Agent Platforms & Tools

🎮 Quiz Activity

Quiz Details

🏆 Quiz Results & Winners

🥇 First Place

🥈 Second Place

🥉 Third Place

📊 Full Results

🔑 Key Insights

💬 Practical Takeaways

📚 Further Learning

🎯 What's Next?