and follow the “Setup before we start” steps if you haven’t already. Thank you!
Introduction
Sébastien Martin
Background
From Paris, France
PhD from MIT
Associate Professor of Operations
Research
Lyft's dispatch algorithm
Waymo Fleet Control
Boston & SF school-bus routing
Teaching at Kellogg
AI Foundations for Managers (AgentOps)
Operations Management
Introduction
Outside of Kellogg
NYSE: ESAB
A global leader in welding & cutting equipment, operating worldwide.
~$2.8B
Annual sales · 2025
120+ yrs
In business
On its Board of Directors
Director since 2026, on the Audit Committee (governance and oversight).
I also advise the company directly on its AI strategy.
Introduction
Our afternoon
1
Hands-on agentic workflow design
we build a real AI agent together in n8n, no experience needed
2
Leveraging AI agents
we practice the decisions behind implementing AI workflows: when is AI the right tool, and where do people stay in charge?
3
AI for personal use
if we have time, 20 minutes on how you can use AI for yourself. Huge things are happening right now.
PART 1 · UNTIL THE BREAK
Building AI workflows with n8n
The best way to understand the technology is to use it.
Building AI workflows with n8n
What is n8n?
An automation platform. You connect blocks into workflows that run on their own, with little or no code.
And one of the most widely used platforms for deploying AI agents in practice.
Building AI workflows with n8n
Skyline Experiences
Premium outdoor experiences, worldwide
Harbour sails and vineyard tours
Hot-air balloons and food tours
Their customers are always travelling, and the weather always matters.
Our goal
A customer-service AI agent for customers who have already booked an experience.
Building AI workflows with n8n
What is an AI agent?
Building AI workflows with n8n
The ONLY three things you control
The model
The LLM, the large language model: the brain. Bigger and smarter, or cheaper and faster. Hosted by a provider, or run on your own machines. Many trade-offs.
The instructions
The system message. Information it was never trained on, and the rules you want it to follow.
The tools
What it can reach and act on. Like you, but with your phone in your hand.
Building AI workflows with n8n
Hands-on 1 · Make it yours
Edit the system message of the chat agent we just built, send a message, see what changes. Try a few, they get harder.
Change the way it speaks: warm, playful, five-star formal, or another language.
Give it facts about Skyline it can share.
Have it upsell another experience when it fits.
Tell it when to use the weather tool, not every time.
Make it always add a 3-day forecast for any city asked.
Swap the model: gpt-5.5-mini → gpt-5.5. Feel the difference.
Building AI workflows with n8n
Hands-on 2 · Give it an inbox
You can just follow along with me, step by step. Lost the thread? Pick up here:
Create your workflow. In n8n, open the CDAIO Build Your Agent project, then add one named after yourself: firstname-lastname-agent2 (e.g. johnsmith-agent2).
Add the template. On this page, copy the code under “2 · Email agent” and paste it onto your workflow canvas.
Set your email. Open the Gmail Trigger node and replace YOUR_EMAIL_HERE@EMAIL.COM with your own address.
Email the agent. From that same address, write to aiml901sebastienmartin@gmail.com, then wait a minute.
Run it by hand. Click each box in order, waiting for the green check before the next.
Go live (optional). Click Publish and the agent runs on its own, checking for mail once a minute.
🎉 Congratulations, you've created an agentic workflow!
Building AI workflows with n8n
Hands-on 3 · The full workflow
Set it up
Same process as the last slide, with a new workflow named after yourself: firstname-lastname-agent3 (e.g. johnsmith-agent3).
This time copy “3 · Human in the loop” from the page and paste it onto the canvas.
Set your email in the Gmail Trigger, as before.
Email the agent, then run each box by hand, or click Publish so it runs automatically on each new email.
Your inbox plays both parts: the customer who writes in, and the reviewer who approves the tricky ones.
Experiment with the AI instructions
Run the whole thing. Try to get one reply the AI sends directly, and one that goes to a human.
Make the router more cautious, so more emails go to a human.
Have the AI always add the customer's local weather forecast to its note for the reviewer.
Building AI workflows with n8n
So why did we do this?
The real takeaway
You don't have to build agents yourself
When a technology moves this fast, what matters is understanding what it can and can't do, enough to put it to work.
The hard questions are business questions, not AI questions. So the real move is to empower your people, more than to hire an outside "AI expert."
But if you're curious
And building one is within reach
In my Kellogg class AIML 901, non-technical students build their own AI agents in five weeks, many now run them at work or started companies.
After every client visit, Proxima's sales representatives lose 1 to 3 hours on paperwork: writing up notes, updating the CRM (the customer database), and drafting follow-up emails.
Who is on it
Maya Chen leads the company's push to put AI agents to work (the AfterVisit AI pilot). She is sure it is a winner. Her ideas may or may not be good.
Your role
You are the advisor Maya brought in. Talk to her and gather as much as you can, she knows a lot. Work out what you would do in her shoes, and how you would use AI.
Leveraging AI agents
Talk to Maya
Maya Chen
Lead, AfterVisit AI Proxima Health Systems
It is an interactive case: you can actually talk to the people in it.
1
Open the class page, then the “Talk to Maya” section. sebastienmartin.info/p/cdaio
2
Interview her. Ask as many questions as you can.
Your goal · figure out
›The situation, deeply: what does Maya want to do?
›Would you do anything differently?
›How exactly would you use AI, and which agent is worth building?
›Where is the line between agents and people?
~15 minthen we debrief together.
Leveraging AI agents
What’s really going on at Proxima
A communication problem between the field and headquarters, and AI is the proposed fix.
Leveraging AI agents
You only met Maya
If you asked, Maya may have shared other people’s points of view across the company.
Maya Chen
Program Lead, AfterVisit AI
The one you talked to
Dana Morales
Senior Sales Rep
“Don’t surveil me”
“I want to stay in control”
Marcus Green
Head of Sales
“Can it connect me and my reps?”
“Will I finally see the field?”
Gavin Holt
Chief Financial Officer
“Prove the ROI”
“What if it fails?”
Morgan Ainsworth
Chief Executive Officer
“Give me a board story”
“Just don’t embarrass us”
Jordan Reeves
Head of IT
“Is our data safe?”
“People use AI in secret”
Leveraging AI agents
One agent, all of this at once
When should a human step in?Will reps feel watched?Where’s the ROI?Who stays in control?Shadow AI is already hereA story for the boardFear of being replacedWhat should the agent do?Data security & privacyWhat to measure in 90 days?Fix the system, not the peopleInclude users in designReplace the costly CRM?Could our data leak?Move fast, or build trust?Give HQ real visibilityRisk of a field revoltKill switch & stage gatesWhere does our data live?Consent to record?
PART 3
Building with AI
AI can do much more for you than run workflows, almost like your own “remote employee.”
Building with AI
What happens if AI keeps getting more powerful?
2022
2023
2024
2025
2026
ChatGPT
AI can write and answer questions
GPT-4 & the AI wave
AI can be really smart
Agentic AI
AI can “reason” and use tools
Claude Code & coding agents
AI can work on its own on complex tasks
Claude Cowork & computer agents
AI can control a computer and do real work
Building with AI
Computer agents
& Cowork
Anthropic
Codex
OpenAI
Whichever you pick, this is the one to get into: an AI agent that works right on your own computer, the closest thing yet to a general-purpose teammate.
Live demo
Let me show you what one of these can actually do.
Building with AI
Why start using computer agents now?
It builds on what you know
There are no AI experts, and you can reach the frontier fast.
AI magnifies the expertise you already have: call it 95% you, 5% AI.
It puts you in control
Use it yourself and you keep your judgment, instead of outsourcing your thinking.
The value sits with the leaders who understand it firsthand, not a hired specialist.
The timing is right
I'm seeing fast adoption at the C-level.
What mattered six months ago barely matters now, so no one is behind.
Building with AI
My own path with AI
What happened when I tried to use AI to change the way I teach.
Building with AI
Thank you!
There are no AI experts. Which is just another way of saying that everyone can become one.
So here's my one piece of advice: build, build, build.
sebastienmartin.info
linkedin.com/in/sebmart
sebastien.martin@kellogg.northwestern.edu
I'll keep the slides and course materials online for a couple of months, in case you'd like to revisit anything.
