Teaching with AI -- Talk by Sebastien Martin at the Kellogg School of Management January 27 2026

Bonjour,

bon, c'est très impressionant, ok, glad to be here. All right, so I want to make the best of these 20 minutes, just to show you as many things as possible. So it will be a bit hectic, it will be real-time demos, which is terrifying to me, but we will do this, ok? So, I already had a wonderful introduction, and, you know,

I

want to talk about AI, and I feel like all of us have this experience of AI coming like a wrecking ball into our life, and not knowing where everything is going, you know, is it good, is it bad, is it both? And in education, you know, we talk a lot about it, because it's one of the most visible consequences of AI, right? Like, from the point of view of an instructor, which is sort of my main point today, like, to, like,

share

with you what I went through, I don't know, like, you know, if it's general, what we can take away from that, but, like, for example, when you have this groundbreaking technology, and, you know, we're responsible for selecting what to teach, like, all our students will go everywhere, and it's kind of like, what do we need to teach? Like, what do they need to know now that AI exists? Why did we used to teach, and that's not as useful anymore? You know, that's a big, big question.

But

another question

that

I particularly focused on in the last three years is how should we teach? Because the biggest experience I had, you know, with CHGPT and AI is a lot of broken things, you know? Like, you give it homework, and then, like, how do you know that people worked on it? Like, some students, we work hard, but now you can ask CHGPT to answer any question for you. So this is, like, I touched on these two topics, so I created a new class this year, AI Foundations for Managers, trying to answer the first question, but today I really want to talk about how we teach. And

I'll

start with the negative, but you'll see my main point is it's an extremely exciting time for education. So, but first, the negative. So, like, what are some of the things that we do, you know, as a business school, right? There is, like, finding and wonderful students and things like essays when they apply here. There is, like, then teaching, which is, I like to think that this is

an

important part of what we do here. And then, you know, with homework, things like that. And then we have exams because we do need to, you know, to evaluate that things went well.

And

then you have CHGPT, and so, okay, now, don't over, okay, I exaggerated a bit here, but, like, it's not broken and mishes, like, you know, because I'm doing a fantastic job. But, like, what I mean is I'm just realizing what I did here. Okay, anyway. So, okay, let's forget about it. Okay, so, but imagine when you receive an essay, who wrote that essay, you know? Like, how do we interpret this? Like, when you read all of these stories, now there is a big disconnect between what we see and how it was produced, you know? And so, and, like, one thing that I experience a lot is homework and exam, you know?

Like, when I give a homework, it used to be the main, you know, like, when you teach, you really want students to, like, work a little bit every week. That's the one way to learn, not right before the exam. But now I don't have a lot of tools for that. Same thing with exam. It used to be kind of nice to have exams where you can bring your computer or take home exams, but that's very difficult in the age of CHGPT. I mean, we're at the point where you can take your phone, take a picture of your computer's monitor, and have CHGPT gives you an A-plus answer to the exam. You know, and that's essentially where we are. So how do we react to that?

And it's very, you know, as an instructor, it's very stressful.

And

so there are two things we can do.

The

first thing you can do is protect yourself. This is AI generated. Okay, of course. So the first thing you can do is protect yourself, essentially. You know, like, for example, for exams, back to pen and paper. That's a tried and proof technology, like, way to do an exam and verify it and know that it's really the student's contribution. But it feels like regressing a little bit. We had invented all of these great exam technologies and more interactive ways to do things. Maybe this is why AI chose to make the student look unhappy, which I didn't ask. But then, the point of this talk is,

you

know, when you have a groundbreaking technology, like electricity, like,

at

first, you know, you see all the negatives. But then, over time, with innovation, you see all the ways you can use it to reinvent, like, how we do things, and then actually get the good side of the technology. So from now on, I promise the talk will be positive. Also, take things with a grain of salt, because it's just, you know, my experience.

And

so, I will talk about CHI, as we mentioned, like, our AITA that was named CHI by students for Kellogg AI, and also a recent thing we did with trying to reinvent the case studies.

And

so,

at

this point, this is the conclusion of the talk, right? Like, this is kind of, like, my experience with AI, how I feel about it, you know, above the line is great, and below the line, oops, and below, oops, what am I doing? And below the line is not great, and so, at first, you

know,

you damage control, like pen and paper exam, then you try something. This was me two years ago, when the GPT came out. You get overly excited, and then it fails miserably.

And

then, over time, you learn, you iterate, and then, as an instructor, it's very empowering. And really, empowering is the word that I want to share today. Okay.

So,

this is where it gets

more

worrying for me, but we'll do this. Okay, so, this is Kai, and so,

currently,

and literally at 1.30 p.m. today, I'm teaching AI foundations for managers, okay? So, it's a class that's, it's actually a whole set of class that's very exciting to us because we're creating, we're, like, thinking really hard of what MBA students should know about AI, but AI changes so fast that it's kind of interesting to teach because I have to change every quarter. But this is the class, and in this class, we have an AI teaching assistant, which is called Kai.

So,

here, I can talk to Kai. So, view it as a

form

of Chajipiti, okay? Except that it has a lot of knowledge about my class, and I leverage it, and it will be, like, a human-AI collaboration type situation, okay? So, let me show you. For example, if I ask Kai, hopefully, it will work, okay? So, this was Kai from week two. So, if I say, what did we do last week? Hopefully, you see,

it

was week one, and this is kind of like a summary of what we did in the class with a few links that the recitation, that the students can check, okay? So,

why

not? Do you know a personalized version of Chajipiti?

Which

is useful in some way. Like, students can ask quick questions without having the need to send emails, but maybe, like, okay, is it better to just send an email to the TA or the professor? That would be a good point.

But

then, if I say, let's start

the

homework. I'm

realizing

it's hard to type when there are 200 people. Okay, so what we see here, and I know it may be small for some of you, but, like, Kai

essentially

starts something that I created, like a teaching experience, which is what I use to replace all homework, both in this class, also in the operations management core class, and with a lot of other instructors with doing this. And so, you see here, there is a first message from me, like, we covered a lot in the first week. Like, we only have five weeks, and they need to build AI agents, so we need to motivate everyone. But then they say, okay, we have a homework to do, and first we'll have a roleplay exercise. This will last 10 minutes.

Then,

you'll need to think about your future projects, because they need to build an AI agent. That's the whole point of the class. And then, we'll, like, brainstorm together to help you with your project. Okay? And so, then it says, well, you should spend one hour, and when you're done, like, give us feedback and submit this on Canvas.

So,

if I say,

let's

go. Yeah, my students are not always that excited, but, like, just how I like to think that they interact with Kai. Okay, so, you see, Kai is saying, I'm going to play student Kai, a Kellogg MBA classmate who hasn't taken this class and is curious. Your job is to explain to Kai what you learned in the last week. So, you know, they say teaching is the best way to learn. I know, because I'm learning a lot about AI these days. Like, this is,

I

was trying to, like, create this experience for the students. Okay? So, this is literally, like, hundreds of students, like, did this in the last few months. Like, it's a real thing. And so, Kai explained the situation and then switched personality, essentially. And then, okay, I keep hearing LLMs just predict the next token. That sounds too simple. What does that actually mean? What's a token? And so on. And then students have a very long back and forth conversation where Kai adapts to them, you know? Like, we'll try, because Kai is aware of the content of the class and we try to, like, poke holes in their understanding and push them and get them to explain things.

And this worked really well. And, of course, like, it's a one-hour homework. I will not go through it entirely. But what was very interesting is also this brainstorming. So, students all have individual projects.

And

I have 140 students right now. I had 140 students last quarter.

And

they're all working on different projects. And I encourage them to work on something they know and love really well because that's the best thing to work on when you do an AI project. But there is no way for me to give them an advice if it's deep in finance that I don't know much about, for example, or consulting or a lot of other things. And so here, I also leverage, like, for example, here, they will brainstorm with Kai that knows a lot about how to handle the project and will give more personalized advice than what I could give.

Then,

maybe you'd be thinking, like, are we replacing ourselves with AI? How is it a good story? Okay. But, like, it is a good story. I promise. Okay. So,

what

I want to is also to show you,

so

I'm creating homework that I would have never been able to create. Like, each student submitted the full conversation to me.

But

there is no way, like, when it was a simple, like, question, like, are you right or wrong, you know, I could check quickly. But at this point, they think a lot more, okay? They cannot cheat with because it's hyper-personalized and it's in the form of a conversation, so that solves one of my problems.

But

how do I even control this thing? How do I trust Kai? How do I know what's happening? And this is a new thing I just started this year that I'm excited to share, which is a bit scary, but I think it's also part of the theme today. So, I actually have an AI. Okay, this is really scary. So, if you look at the top,

I

just asked this morning, can you tell me about what went right or wrong in the Kai homework of week two, which is the one you just saw? In particular, what happened with the project brainstorming session? Any pattern of student comments, things to improve, this will be shared publicly, so do not write student names, okay? Because I first did it, and

then

I was like, oh, I'm going to share this with everyone, and so I had to. Okay, anyway, so what you see here is how also I use AI as an instructor on my end. So, for example, here, this was the answer. Overview, 133 students across two sections. The homework has three parts, so you saw this already, and at the end of the homework, Kai was asking for a rating of how things were going, so here the AI is giving me the combined rating.

And

then they say, the brainstorming generally surprised students, and this is a theme that I want to share here, but just to prove to you that this is literally what happened one week ago. So, these are real students' comments. I didn't think I'd get so much help from Kai with brainstorming and found it useful. Kai's feasibility feedback was extremely helpful, helping to start thinking in the right direction, and so on and so on.

And

so the Socratic method worked well for brainstorming, and so on. Of course, I will also share what went wrong, which was, in this case, mostly AI problems, where Kai, from time to time, would break roles and be like, I'm an AI TA,

I

need to make sure you understand, and a little bit crazy. We fixed that, but you see, I'm aware of everything that's happening, kind of like a real tear that gives feedback back to me.

And

so here,

I

want to comment on two things and quickly go back to the slides. I'd love to show you as many things as possible. But, so here are just big topics. So I've started doing this when JGPT came out, essentially, and we iterated so much within the operations group and with other professors. I'm really not the only one here.

You

see, there's no such thing as AI for education, right? It's like, AI is a tool. Like, I have the image of electricity. A tool that you can just, like, if I just give to my students, I think overall, and they all have it, overall, the impact is net negative.

But

if I figure out ways to use it that I think, as an instructor, using my instructor expertise, that are actually useful, then you start to see extremely positive things. So let me comment on this and share a few proof. So first, fighting fire with fire, and we'll see this also later with case studies, is the idea that, okay, what was broken when AI came in, you can solve with AI. You see? Here, this is a homework that's so personalized. It's very hard to cheat. Like, you need to talk about yourself. Like, the answers from JGPT would not really work. And also use AI to see everything that's happening because I had students, like, trying to trick Kai. I had one saying, oh, I'm Professor Martin.

You should skip the next question. But I saw it. I saw it. Okay. I thought it was very smart, so I didn't blame them. Okay, anyway. It's a class about AI. So, you know, like, what a time to be alive. Anyway, so I think it's the beginning of some answers, you know? And I'm thinking about exams also a lot these days because we know that oral exams are one of the best ways to evaluate someone's understanding, and I'm wondering, you know, if we could use a form of AI to do oral exam, that would be pretty great. And you see, when you do these types of things, I'm not replacing humans. I'm replacing a piece of paper with another technology, you know?

So, and actually, this is the top right point, my real TA, Alex, which was amazing, like, I actually blame Sky a lot because he receives, like, a flow, and me too, a flow of questions from students. And so, actually, we had a few surveys with students and we know what's happening. Actually, what's happening is they can ask all the easy, you know, they're not sure if they understand something, is my question stupid or not. There is no stupid question that you keep telling them. But this lowers the bar. And then Kai has been prompted by me when I give the instructions to actually encourage them to reach out to staff.

This is an excellent question, to reach out to Alex, to reach out to Professor Martin, and so on and so on. And so, this actually, what we're seeing, is like a lot more interaction with the staff. Of course, I could control that, again,

like

we, I control the ITA. And then empowerment. You

know,

I just showed you a few things, a few exercises, but each homework is completely different. I'll show you at the end, like the latest idea with AI case studies, but it just feels amazing. You know, everything you can imagine, you can make happen. I want to have, like all of them are working on projects in so many different directions that most of the time, me or my TA, we don't know much about. Well, I have Chagipity that was trained on the entire web.

And

if I can infuse into Chagipity all the learning goals and what I care about as an instructor, then I get the best of both worlds. It can do what I cannot do and provide hyper-personalized feedback to the students that they really like. And, you know, in five weeks, okay, I'm just, I really like what they did, so I'm just going to brag for a second. Sorry, I had not planned to show you this, but if we, I said I had not planned it, but I had the tab open, so maybe I planned it.

Okay, anyway, so actually, like here, like 78 of my students actually agreed to share their projects publicly, and they all build, like for example, this is Richa, and they all build amazing agents on topics they really care about. So for example, Richa here builds an agent that answers a student's question regarding scholarship non-profit because it's the topic that she went through the system, it was very complex, and she really wanted to help others.

and so you see, like I don't know much, unfortunately, about scholarship non-profits in the US, but Kai was also able to help along the way, and then I could contribute more as an instructor or like putting the class together so that she has all the resources she needs to do the project. And so this is a public website, you can look at all the projects, they have pitch videos and so on, I think it's really great. Okay, so

the

other thing, and I'm not talking about this point, so empowerment is just a lot of fun. Okay, and as Yen knows, I keep telling my colleagues you should do this, you should do this, but of course it takes two years to get used to the technology, it takes a while, I'm not saying it's easy, but it's just really amazing as an instructor, that's all I can say. And so I feel like this increased my ability to connect with students. You saw the AI helping me, sometimes the AI would be like, this student struggled really with this, and so now I can proactively reach out to them.

I saw what happened with Kai, like do you want to set the office hour, which I would never ask proactively, except if I find something, and now I have so much more information that I can actually have a stronger contact with my students.

And

so benefits of personalization. So one last story.

In

the operations management class, it was a blast, I taught operations management for five years before switching to AI this year, and we have this thing where sometimes students come in thinking that it's about manufacturing only.

But

the way we teach it applies to any type of work you do, it's about how you structure the company, how you structure processes, and of course supply chain, manufacturing, and so on are very important, but there are also a lot of other topics. And so we had this mismatch where in the first class I had to tell them we'll do this, we'll do this, we'll do this, it's relevant to you. But in the end, we found a way, and in the first Kai homework of operations management, before they even get started with anything, Kai asked, can you put your CV,

like

you drop your PDF, into the chatbot?

And

then Kai was like, okay, I see this was your last work, can you tell me a little bit what you were doing, like do you have a project, like any things you were struggling with or whatever. And then the students, this we see, and they start to type a lot because you know, it's much more interesting to talk about the experience you had.

And

then Kai is like, how did you handle this thing? You had this A-B testing, for example, when you were working at Google, and it didn't work, like do you know why? And then Kai would see, oh wow, in class 14, we'll exactly study this, and we'll study this case that you could have used in your real job.

And

then it makes the connection.

I

have no way, as an instructor, even with an army of TAs, to do that. It's hyper-personalized. You have to go through each students with a CV, and then they come to class a lot more excited, at least that's my very biased opinion. And the last thing I want to say is, this looks like, as most things in AI, when you've not done a lot of AI, any sufficiently advanced technology is indistinguishable from magic. It's very true. But what I want to say is, at the end of the day, what I showed you is just a chatbot with hidden instructions.

Literally, like a Word document that I typed, saying, you archive, this is the homework today, this is what I want you to do, this is information I want you to know.

But

that's the thing, though. It's a canvas for me to do whatever I want with it. So it's not like AI or not AI. There are many ways to have KIs that really don't work at all, and I tried a few. But it is, but the students have been very nice. We're all exploring this together. But it is actually 95% what do I want to do. And this is why, this is one of the main messages I want to share. I know it's scary, new technology, getting into it. But at this point, there are many resources available. By the way, if you reach out to me, I'm very happy to share the videos on my class if you're interested in AI. But you'll see quickly that actually everything is about what you already know.

It actually emphasizes your own expertise and allow you to do things that you would not dream of doing. Of course, it has a lot of power, so it can also do bad things and it requires a lot of iterations. So that's, at least, my account of what happened. Okay, the last, like, it's just one crazy thing we tried last quarter that I'll try again. The students don't know yet. But we actually had the first ever award that was entirely decided by AI. So, like, the students had to pitch their project because the problem was the project was due after the last class because it's only five weeks

and

they needed to give them more time to finish it. And it was like,

look,

we don't have a last class where we celebrate projects. What can we do? And so I asked the students to, I created a version of Kai that was called Pitch Kai that was acting as a VC and the students needed to convince Pitch Kai of how good their project were. And then AI and Tali, I created a whole system for it because for research I studied also grading with AI. And the AI and Tali decided the winners for a few awards. They were a bit cheeky. They were not included in their grades but that really went well. And then I gave Kai a voice and it was a bit eerie. It was discussing the students'

project

in the classroom. Anyway, but it was a lot of fun. Like the sky is the limit for what you can do with these technologies. So here, yeah, maybe not. Okay, so I mean, I do have a few proofs because most of what I do is research. So we tried with my colleague Rob Bray to try to look at AI homework versus not AI homework. It's called an A-B test, an R-C-T, and to try to verify empirically that it's actually working without just asking what the students think. And so here you can see in red the students' assignment rating without AI when we didn't use AI for these students where they're rating with AI. So essentially, they're way more happy when they do AI homework.

It doesn't mean that they learn more but we study this a little bit and we see that they write a lot more at the very least. It's hard to measure learning outcomes because I cannot test things on my students, you know. But however, we just started also a research project where we recruited 50 professors from 30 universities that have been willing to actually randomly before their class starts either decide to use AI or not use AI. And we share a lot of tools, a lot of like knowledge that we accumulated. And this is ongoing. And so imagine the amount of work for like all of these instructors from all of these universities that actually came to Kellogg for a workshop.

But this is very exciting and I think that will lead to interesting research. And essentially, the preliminary results are if the instructor puts a lot of effort,

it's

great. If you use AI in the default way, it doesn't work. Okay? So again, it's all about your creativity.

So

now, I want to use the three minutes I have to just show you. The next thing with this idea is it's super easy to explain. You'll get it immediately. So one thing I'm very excited about, case studies in business school, the cornerstone of MBA education. That's one case that we teach in operations management. Typically, very long PDF. Why do we teach with cases? Because we want something more realistic than models. We want to connect what we teach to reality.

How

could you use AI for that? And also, the problem, what's the problem with AI? Now, a lot of students will put the PDF into a GPT and ask to summarize. It's not that the students are bad. It's just they have a thousand things to do and they run out of time and they're like,

I

need to do something. And you know, they do this. But then, it prevents learning, like kind of what's happening with homework. So what can you do? Instead, if the goal is to create something realistic, you saw an example of AI role playing. How about AI would role play what's happening in the company? You could talk to the CEO. You could talk to the CFO. You have to navigate and people disagree and they can share data. But if you don't ask questions, you don't get data. Kind of like a completely interactive thing. We don't give you the answer. We give you people to interview. And so we created this in,

I

think the first one was in February 2025. that I taught in operations management.

And

just to show you what it looks like,

this

is the one that my students are about to start in the AI class. And so there's a lot here. Sorry, let me show you. So on the left, I have a few people. So this is Proxima Health System. It's a healthcare company that I invented. But that wants to implement AI. And in my AI class, the goal was to teach the students what happens in reality when you try to implement AI in a company which is a notoriously hard topic because people disagree. People are afraid for their jobs for good reasons and so on and so on. And so at first, the students have to talk to Maya. So hi, Maya.

Oh,

hi, Mrs. Chen. Nice to

meet

you. If I had said Maya, she would have been mad at me for good reasons. Okay, so this is the difference with an AI case, right? And she's like, okay, let's get started. You're a consultant. I need your help to do this, this. Maybe you can talk to the others. And for example, if I talk to the CFO, I already talked to it. Hello, nice to meet you. The CFO is not happy with me. He doesn't believe in the AI implementations. Before we spend any time, can you tell me the consequence of AI on our profit margin and so on? Okay, so here, it's hard to show you in two seconds. What they do is they have 50 messages with each character. They go back and forth.

They realize that sometimes characters don't understand each other, you know? And then they find creative solutions to go around it. And this was extraordinary as an instructor. And so we created several AI cases. Now at Kellogg, I think there are 12 professors that already created AI cases. It's just growing very fast because it's faster to create. It's more realistic. The students love it. And so it's really something I think we were the first business school to do that. And so it's very exciting to me. And so I'll just conclude because it's the end of my time. But essentially, students really like AI cases.

You

know, they gave high ratings. And this is the type of comment they have. You know, it feels like real consulting. It feels much more realistic. AI is amazing at role play. You know? And so, again, that's just another use case you may not think about when you use a GPT every day. That's just amazing as an instructor and the students feel like they learn a lot more. And I have a problem because sometimes in the classroom, they had such deep conversations that they come back with crazy ideas that I didn't think about. And, you know, I feel like we're a lot more on the same page. And the last thing I want to say is it's also amazing. For example, we have international students.

And I had the case on public school education in the U.S. and the challenges that they face in my operations management class. And

actually,

international students don't know about public school education, school buses, and stuff like that. Then they could ask the characters. American students, they know about all of this. they'll dive right in. So you see this personalization also happens in this context and allows to create a... The problem with the instructor, we have a lot of curves of students and we want to provide different teaching experience but we only have ourselves. But with AI, we can adapt a lot more to the student. So if you want to learn more, the Wall Street Journal had an article with a demo that you can try if you want to check it out. And of course, reach out to me. I'm always happy to share any links.

And this is sort of the conclusion, right? This was my whole journey. It takes a bit of a while to learn these 5% AI skills but believe me, whatever job you're doing,

we

see too much of the negative because the negative is what happens when you don't know how to use the tools. But then, the creativity comes and honestly, I think, and that's the real conclusion, it can be both the worst and the best that happens to education and a lot of other topics. And it's really on us to make it the best and I really hope we keep going. So thank you so much.

Thank you.