Northwestern|Kellogg

AI in Education Workshop

Welcome! · Kellogg School of Management · July 9–10, 2026
Today · Thursday, July 9
12:00–12:30
Welcome & lunch
12:30–1:30
Speed Talks: State of AI in Education across Universities · Sébastien Martin
2:00–3:00
Paper + Results & Publication Strategy · Rob Bray
3:30–5:00
AI in Teaching Lab · Sébastien Martin
5:30–6:30
Group Discussion 1: Innovating with AI in the Classroom · Jun Li
7:00
Conference dinner at The Barn Steakhouse
You are here
Global Hub, Room 4101. All sessions in this room, with coffee breaks in between.
Dinner, 7:00 PM
The Barn Steakhouse · 1016 Church St, Evanston (entrance in the rear alley).
Tomorrow
Breakfast from 8:30 AM · sessions 9:00–12:30 · lunch & farewell, done by 1:00 PM.
Northwestern|Kellogg

Speed Talks

Short, informal updates from each school on the state of AI in education.
AI in Education WorkshopKellogg School of Management
Thursday, July 9, 2026 · 12:30–1:30 PM
Speed Talks

Today’s speakers

Christian Blanco
Ohio State
Mi Kyong Wilson
Ohio State
Pnina Feldman
UVA Darden
Arthur Delarue
UVA Darden
Zonghao Yang
Stevens
Wenchang Zhang
Indiana Kelley
Ben Collier
Carnegie Mellon
Ken Moon
Cornell
Blair Flicker
South Carolina
Minje Park
HKU
Jun Li
Michigan
Rob Bray
Northwestern
Northwestern|Kellogg

AI in Teaching
Lab

A live tour of AI in my teaching, and of the computer agents that build it.
Sébastien MartinAssociate Professor of Operations · Northwestern Kellogg
AI in Education Workshop · July 9, 2026 · 3:30–5:00 PM
QR code to the slides and supporting material
Slides & materials
sebastienmartin.info/p/kaihw.html
Introduction

Navigating a technology disruption

Students writing by hand, phones banned
Damage controlThe new technology hurts our old way of doing things, and we try to work around it.
OpenAI vs Kellogg robots with flamethrowers
Build something newOr we use that same technology to do things we could never do before.
Introduction

Our 90 minutes

1
The many ways to use AI in teaching
~45 min
2
Leveraging computer agents
~45 min
Show you what is possible when you go all in on AI —
and the path to get there.
Student-facing AI

Student-facing AI

What my students see: an AI tutor in every course, and cases they can talk to.
Student-facing AI
Live demo

Using an AI tutor

Kai, Kellogg’s AI TA

  • 24/7 Q&A on the course material.
  • AI homework: interactive, personalized assignments, hard to plagiarize by design.
  • Reports back to the instructor: misconceptions, engagement, who is lost.
QR code to try the real week-2 Kai homework
Try the real week‑2 homework yourself
Student-facing AI

Does AI homework work? A mini-RCT

The experiment

  • Fall 2024: 19 MBA homeworks, randomized per student: custom AI tutor vs. base ChatGPT (same GPT-4o, same problems). 139 students, ~27k questions answered.
  • Students prefer the AI tutor, and more so on harder assignments.
  • Honest nulls: quiz scores, time spent, completion unchanged.
Share of homework submissions rated 5/5 Base ChatGPT Custom AI tutor 0 10 20 30 40% 26.5% 39.9% 29.4% 38.3% “Rate your experience” p ≈ 10⁻²⁵ “Helpful for quiz prep?” p ≈ 10⁻¹²
AI homework increases student satisfaction, Bray & Martin (major revision at INFORMS Transactions on Education).
Student-facing AI
Live demo

AI cases you can talk to

The Wall Street Journal: AI Is Teaching the Next Generation of M.B.A.s the Classic Case Study
QR code to the WSJ article and its live AI-case demo
wsj.com/tech/ai/…the-classic-case-study
Students interview the characters of a real case. The QR opens the article and a live demo anyone can try.
Student-facing AI

What AI adds to a case

It adapts
To each student and each goal. An international student can ask what a US school district is; another dives straight in.
It feels real
Students practice the human side: communication, personalities, pressure.
Ask the right question
Real problems have no guidebook. Students learn to figure out what they need to know, and how to get it.
The tutor is built in
Characters can share data, software, or documents, and walk students through them.
Better as AI improves
Voice is already arriving. Every model upgrade makes the characters, and the case, better.
Empower the instructor
Building and changing even complex simulations is easy now. You stay in control.
Professor-facing AI

Professor-facing AI

My latest experiments.
Professor-facing AI

AI homework needs AI evaluation

AI homework

  • Personalized: every submission is different.
  • A great learning tool.
  • Nearly impossible to grade by hand.

AI evaluation

  • Grades and rich feedback for every student.
  • Surfaces the important issues to the instructor.
  • Makes the homework itself better.
Professor-facing AI

How to do AI evaluation?

Idea 1: Homework Grading rubric LLM Grade / feedback noisy, biased, hard to trust Idea 2: Homework of student A Homework of student B Evaluation criteria LLM Which one is better, and why? Grades rebuilt from many comparisons (Elo scores)
LLMs are unreliable graders but reliable comparators: ask which is better, many times.
Professor-facing AI

Early tests: AI-judged awards

Students pitch their final project to the AI
▼
The AI grades and picks the award winners
▼
The AI gives feedback for students to improve
“The time unlocked and depth of feedback is unmatched.”
The Future Unicorn Award plaque, chosen by Kai, Kellogg's AI teaching assistant
Professor-facing AI

This year: grading a whole course with AI

Winter 2026 140 MBA final projects AI pairwise comparisons on each criterion Elo-style scores, feedback, flagged issues Instructor review app: I decide the grades video, transcript, AI summary: all 140 reviewed, much faster
The AI surfaces and ranks; the instructor decides.
Teaching AI

Teaching AI: what my MBAs built in five weeks

Track medication adherence by texting patients personalized reminders
Track competitor clinical trials to inform pharma strategy
Answer customer questions automatically and route complex ones to experts
QR to the AIML 901 syllabus and projects
AIML 901 “AgentOps”
No coding background · 193 projects
Turn course syllabi into study plans with deadlines
Sort and prioritize a travel agency’s inbox, flagging urgent requests
Monitor news and alert retail investors when their stocks are at risk
Teaching AI

My experience with AI in education

AI impact Time Before AI Damage control First steps Innovation
Four years in, the one thing that predicts success: how involved the instructor is.
Leveraging computer agents

Leveraging computer agents

In my opinion, the most empowering way to use AI as an instructor.
Leveraging computer agents

From ChatGPT to Claude Cowork

2022
2023
2024
2025
2026
ChatGPT
LLMs can write and answer questions
GPT-4 & the AI wave
LLMs can be really smart
Agentic AI
LLMs can “reason” and use tools
Claude Code & coding agents
LLMs can work independently on complex tasks
Claude Cowork & computer agents
LLMs can control a computer and do real work
Leveraging computer agents

Computer agents do the work

A chatbot talks about your work. A computer agent does the work, on your files and your computer.
Live: Claude Cowork  ·  Claude Code
Leveraging computer agents
Live demos

What can a professor hand to an agent?

Grade a stack of homework
Help you make slides
Build an interactive demo
Analyze course feedback
Draft AI homework & rubrics
Create & iterate on AI prompts
Find the student who needs you
Kill the boring parts
Just a few examples — we’ll do some of these live, right now.
Leveraging computer agents

Getting started

  • No coding required. Talking is the interface; iterating is the skill.
  • Pick one real task from your week, and give the AI your actual files.
  • Start with Claude Cowork (desktop app), graduate to Claude Code (terminal). Codex on the OpenAI side.
  • Cost: roughly $20–200/month, typically reimbursable from research accounts.
  • Build, build, build. Instructor involvement is the whole game.
Thank you
sebastienmartin.info
sebastienmartin.info
WSJ article and live AI-case demo
WSJ & live demo
AI class syllabus
AI class syllabus