In the previous two lessons you learned what an AI agent is and how it is built. This lesson shows you real agents running in real systems. All three demos follow the same pattern: multiple AI agents working together, each with a specific role, coordinated by an orchestrator that decides which agent handles which part of the task.
As you watch each demo, use the five-component framework from Lesson 1.2a to identify what each agent is perceiving, planning, using, remembering, and checking. The goal is to train yourself to read what is happening underneath — not just to watch what the agent produces.
1
Weather + Travel Suitability Agents Two agents collaborate inside an agentic workflow. One fetches weather data. The other evaluates whether conditions are suitable for travel. An orchestrator decides which agent runs when.
2
Multi-Agent Workflow in n8n Multiple agents with assigned roles work together in n8n. Shows how an orchestrator delegates subtasks to specialist agents and assembles the final result.
3
Agentic AI with Multiple LLMs Multiple agents powered by different LLMs collaborate step by step. Shows how agentic AI scales beyond a single model into a coordinated system.
Demo 1 — Weather Agent + Travel Suitability Agent
Two AI agents are built separately, then combined into a single agentic workflow. The first agent fetches live weather data for any city. The second agent takes that weather data and decides whether conditions are suitable for travel based on temperature and humidity thresholds. An orchestrator coordinates which agent runs and passes outputs between them.
Watch: Two AI agents working together inside an orchestrated agentic workflow (13 min)
What to Watch For
Notice the moment the orchestrator decides which agent to trigger — that is planning at the system level, not the agent level. Also notice how the output of Agent 1 becomes the input of Agent 2. This handoff is what makes it a multi-agent system rather than a single agent with two tools.
Component Breakdown
Perception
Agent 1 receives a city name as input and calls a live weather API. Agent 2 receives the weather JSON output from Agent 1.
Planning
The orchestrator decides which agent to trigger first and routes Agent 1's output to Agent 2 automatically.
Tool Use
Agent 1 calls the OpenWeatherMap API. Agent 2 runs a condition-checking script against temperature and humidity thresholds.
Memory
Each agent holds its inputs and outputs in context. The orchestrator maintains the full conversation state across both agents.
Feedback Loop
The system runs in supervised mode first — asking for human approval before triggering each agent. Then switched to autonomous — both agents run without intervention.
Key Insight
The switch from supervised to autonomous mode is one of the most important concepts in agentic AI. Supervised means a human approves each action. Autonomous means the agent acts without waiting. Most production systems start supervised and move to autonomous once the agent's behaviour is verified.
Demo 2 — Multi-Agent Workflow in n8n
This demo shows multiple agents built in n8n — the same tool Track A students will use in Lesson 4.1. Each agent has a defined role. An orchestrator node routes tasks to the right agent based on the input. Watch how the no-code environment makes the agent architecture visible — you can see exactly what each agent is doing and when.
Watch: Multi-agent workflow in n8n — orchestrator routes tasks to specialist agents (20 min)
Track A Note
This demo is built in n8n — exactly the tool you will use in Lesson 4.1. If the canvas looks complex now, it will look completely readable after Week 4. For now, focus on identifying the five components rather than the technical setup.
Component Breakdown
Perception
Each agent receives a structured input from the orchestrator — not raw user text, but a processed task definition.
Planning
The orchestrator node reads the incoming request and decides which specialist agent should handle it.
Tool Use
Each agent has its own tool set. One might call an API. Another might query a database. The orchestrator itself calls no external tools — it only routes.
Memory
n8n maintains the full execution context across the workflow — each agent can see what previous agents produced.
Feedback Loop
The orchestrator checks the output of each agent before passing it to the next. If a step fails, the workflow stops and flags the failure rather than continuing silently.
Demo 3 — Agentic AI with Multiple LLMs
This demo shows how agentic AI scales: instead of one model doing everything, different LLMs are assigned to different parts of the task based on their strengths. A reasoning model handles planning. A faster model handles data retrieval. The outputs are assembled into a final result. This is the architecture behind many real-world AI products.
Watch: Multiple LLMs working as agents together — how orchestration scales (18 min)
What to Watch For
Notice how no single model does everything. The division of labour between models is itself a design decision — and it directly affects cost, speed, and quality. This is the kind of architectural thinking you will apply in Lesson 5.1.
Component Breakdown
Perception
Each agent receives a specific, scoped input — not the full original request. The orchestrator preprocesses the input before routing it.
Planning
The orchestrator decides which LLM handles which subtask. This is itself a planning operation, often performed by a separate reasoning model.
Tool Use
Different agents call different tools. One searches the web. Another reads documents. Another generates a summary. Tools are assigned per agent, not shared globally.
Memory
A shared memory layer allows agents to read each other's outputs without the orchestrator having to relay everything manually.
Feedback Loop
Before the final output is delivered, a review agent checks it against the original request. This is a verification step at the system level — above any individual agent.
Key Insight
A single agent is a feature. A coordinated team of agents is a system. The orchestrator is what turns a collection of agents into something that can solve complex, multi-step problems reliably.
After Watching — Reading a Multi-Agent System
You have now seen three multi-agent systems in action. Before moving on, try this: pick one of the three demos and map the entire system — not just one agent — onto the five components. Where does perception happen? At the agent level, the orchestrator level, or both? Where does the feedback loop sit — inside an agent, or above all of them?
These are the questions you will answer for your own project in Lesson 5.1. Starting to think about them now gives you a significant advantage when architecture design begins.
✏️ Student Reflection
Pick one of the three demos. Describe what the orchestrator is doing — not what the individual agents are doing, but what the orchestrator specifically decides and coordinates. Then describe one real-world scenario from your own domain where a multi-agent system with an orchestrator would be more powerful than a single agent alone. Be specific about what each agent would do and why one agent could not do it all.
📝 Your Lesson Submission
Share your reflection, ask a question, or leave a comment. Your mentor will respond.
🔐
You need to be logged in to submit your reflection.
Next →2.1 — AI Ethics and Responsibility🔒 Submit reflection to unlock
🔐
Create a Free Account to Continue
Lesson 1.1 and beyond are available to enrolled members. Join the program for free to access all 21+ lessons, mentor sessions, and your completion certificate.