---
title: "The Agentic Classroom"
author: "Geoffrey Challen"
date: "2026-11-20"
description: "A guest talk for the Osher Lifelong Learning Institute at the University of Pittsburgh: what an AI agent is and how it differs from a chatbot, three classrooms where agents bring students into conversation with each other, and what students say they want from their university."
canonical: "https://www.geoffreychallen.com/talks/2026-11-20-the-agentic-classroom"
audience: "Osher Lifelong Learning Institute, University of Pittsburgh (Zoom)"
slides: "https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html"
film: "https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/film.mp4?v=b57e340e25"
---

# The Agentic Classroom

[View the slides](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html)

[Watch the film](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/film.mp4?v=b57e340e25)

## Summary

Computing in human language is a dream as old as computing.
Grace Hopper wanted people to work with computers in ways that were natural for humans, and for seventy years we made people learn the machine’s language instead.
Today her dream is coming true.

Most people have met AI as a chatbot: an assistant stuck behind glass that can only give advice while you do the work.
An AI agent can act.
It writes the code, searches the project, and does the work alongside you.
About a billion people use a chatbot each week, and only a few million work with an agent, so there is a growing gap between what most people think AI is and what it can do.
The printing press dates from the 1400s and mass literacy took centuries; I hope this takes less time, and educators have a role in that.

I described three classrooms.
[Using and Understanding AI](https://usingandunderstanding.ai) is a course about AI for any student, with no lectures: students explore, discuss, and build.
A handwritten-digit recognizer they can train in the browser took me a couple of hours to build with an agent.
Before class, students talk through the reading with an agent so that they arrive as better partners; in class, agents help form groups and guide discussion; and assessment can be a conversation in which a student explains an idea and then defends a position against pushback.
In [CS 124](https://www.cs124.org/ai), about a thousand students meet every Tuesday for an activity rather than a lecture, and last spring’s [Duckies](https://www.cs124.org/duckies) included a whole category of apps built for older relatives.
In [Conversational Programming](https://www.conversationalprogramming.org/educators), nineteen students with no prerequisites build with agents and never touch code: their own website, something for a classmate, and now apps to help students find connection on campus.
In all three the aim is the same: use AI to bring students into contact with each other, not to pull them into their laptops.

Working with AI is a skill, not magic.
We should teach it early and often, AI can help us teach it, and students enjoy learning it.
It also points to a different future for technology: not a few people building what everyone has to use, but many people building for themselves and for the communities they care about.

I ended with something from that week.
In CS 124’s activity on AI and education, 436 students each made one ask of the university, and the most common was: teach us how to learn with AI.

## Abstract as Announced

Computing in human language is a dream as old as computing.
Today it’s coming true all around us.
Good coding agents are barely a year old, and already people with no technical training are using them to create technology.
And we haven’t even started teaching people how to do this yet!
This fall I launched a new course on conversational programming, with no prerequisites, in which students build websites, tools, games, and analyses with an AI coding agent, all without ever interacting with code.
I’ll describe how the course works and what we’ve learned so far about teaching people to collaborate with AI.
I’ll also discuss how we’re using AI to support student learning, both inside and outside computing courses.

## Slides

### Why I am here

[![Slide 1: Title](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/01-title.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/0)

Thanks, everybody. I was really pleased to be invited. Cory is a good friend, and he and I have done some fun things together. I could talk about this topic for several hours and I have twenty minutes, so I am not going to try to cover everything we are doing.

### A dream as old as computing

[![Slide 2: A dream as old as computing](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/02-two-dreams.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/1)

This is an exciting moment for computing. I am a computing educator, and ever since the field started we have had this dream that it would be easier for people to work with computers.

Grace Hopper, an early computing pioneer, wanted people to be able to communicate with computers in ways that were more natural for humans. Unfortunately, seventy years went by during which the primary way we were forced to communicate with computers was through computer language, which is slow, unpleasant and somewhat dehumanizing. What we are seeing happen all around us now is Grace Hopper’s dream coming true. I am sad that she is not alive to see it, because I think she would be pretty excited about what is taking place.

### What an AI agent is

[![Slide 3: A chatbot and an agent](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/03-chatbot-vs-agent.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/2)

There are lots of different ways that people are interacting with AI right now. Many of you have probably used an AI chatbot, maybe through a web browser. I think of that as your AI stuck inside a constraint. It is trapped in the web browser, unable to help you with certain things in a direct way.

What I am talking about today is working with an AI agent. As the name implies, these are tools that have agency. They can do things. With a chatbot you are in charge of doing whatever it asks you to do. An AI agent has much more ability to affect its environment: it can write code, search the project, and take actions on its own.

[![Slide 4: A New Skill](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/04-a-year-old.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/3)

These are the tools that are causing a small group of people to get very excited about this technology, and it is still a very small group. Andrej Karpathy recently did some back-of-the-envelope calculations about how many people are actually using AI.

People interacting with a chatbot: about a billion a week.

There is a much, much smaller group using AI agents, with the latest and most capable models.

I drew that on a log scale. Put it on an absolute scale and there is a growing gap, and I hear it in how people talk about AI: between people using it as a chatbot, sometimes with limited success, and people using the more powerful versions as agents. The rest of this talk is about teaching students to work with AI agents, not just chatbots.

### The printing press

[![Slide 5: How Long Reading Took](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/05-literacy.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/4)

I have been thinking a lot about historical precedents for this technology. It is striking to me that the invention that really enabled written culture, the printing press, dates from the 1400s.

It was hundreds of years before mass literacy started to emerge across the world. Some historians point out that it was to some degree the Protestant Reformation that created an interest in literacy, because for the first time people were told they should meet the word of God in written form.

I am hoping that with these tools we do not have to wait centuries for large numbers of people to have access to this technology and to use it to improve the world around them. I think educators have a real role to play.

### A course about AI, for everyone

[![Slide 6: The other dream](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/06-two-dreams-again.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/5)

That is why there is a unique opportunity right now for computing educators. We are seeing the changes these tools bring to computing.

And we are at a point in time where really positive changes are coming to education from the same AI agents: the other old dream, a tutor for everyone. Those are things I am doing in my own classrooms.

[![Slide 7: Using and Understanding AI](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/07-uai-goals.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/6)

I will talk briefly about a few courses. This is one I taught last spring, Using and Understanding AI. It is what is sometimes called an AI literacy course: a broad introduction to what AI can do, the impact it is having on our society, and the basics of how it works.

[![Slide 8: Understanding AI Lessons](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/08-uai-meetings.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/7)

There are three types of meetings. In some, students talk to each other. In others they do something with AI. In others they learn something about AI. All of them are group work. I have long since abandoned any interest in lecturing. When students are in my classroom I want them working together, in conversation with each other, because that is the best way they learn.

[![Slide 9: Inductive Exploration: Digit Recognition](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/09-inductive-exploration.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/8)

Here is an example of the things you can create today to support a course like this. At some point we teach students a little about neural networks, the basis of generative AI. This is a small network that recognizes handwritten digits. You can draw a digit and see the probabilities, and you can train the model right in the browser and watch the weights between the nodes change as it learns. It gives students a visceral sense of how these tools work. I built it in a couple of hours with an AI agent. I never would have attempted it before, because it would have taken me months to build one visualization that we used in one class for half an hour.

[![Slide 10: Conversational Preparation and Engagement](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/10-uai-prep.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/9)

One of my big focuses is using AI to help students interact with each other. I do not want students to disappear into cyberspace. I want to use AI to bring them into human connection.

In this class students talk to each other about the readings, and they do not always show up prepared. So before class they have a short chat with an AI agent about the reading, to make sure they have the key ideas and to make them a better partner. If you did the reading and you are paired with someone who did not, that is an unpleasant experience, and you start to wonder why you did the reading.

In class, agents help organize students into groups, suggest who they might want to talk to, and help guide the conversations. I am happiest when I am in a classroom and there is a din of students talking to each other.

[![Slide 11: Conversational Assessment](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/11-key-idea.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/10)

Another thing we are doing I call conversational assessment. How do you know that a student knows a topic? We have used blue book exams and multiple choice questions for years. There is a new opportunity here: establish what a student knows by having them talk to AI.

This one asks whether the student understands the Turing test. That is a factual question, and it is the first thing the agent establishes.

The second part is more interesting. It asks the student to take a position: do current AI models pass the Turing test? A student can take either side, and then they have to defend it. The agent pushes back with counterarguments, and how well they do depends on how well they engage with them.

### A thousand students, fourteen Tuesdays

[![Slide 12: Fourteen Tuesdays](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/12-tuesday-activities.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/11)

This is a much larger course, about a thousand students, where we are also teaching them to use AI to create technology. They come together on Tuesdays for an activity, and we use AI to structure those activities, again to give students the chance to work together and talk to each other. We ask students to come to campus and into a classroom. Lecturing does not value that time. When I bring students into a room, I want them doing something they cannot do alone in their dorm room.

[![Slide 13: The Duckies](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/13-showcase-duckies.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/12)

As we have given students the chance to use AI to create things, they have done incredible projects. We started some awards in my introductory course, and we will do them again this semester, to recognize students who built really remarkable things.

[![Slide 14: For real users](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/14-for-real-users.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/13)

One category that emerged last semester seemed relevant to this group: quite a few students built apps for older relatives, to help them use their devices more effectively. When we give students the opportunity to create technology, they find really meaningful things to do with it, to help the people around them.

### A course with no code

[![Slide 15: The course](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/15-the-course.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/14)

The last course is one I am teaching right now, inspired by and designed along with Cory. In my traditional programming course I still teach students to write code by hand, but I think the days of people doing that are numbered. In this course we give students AI agents and see how much they can do without ever working with the code.

The first thing they did was build their own website.

Then they built something for a classmate. People came up with ideas, and students built for somebody else, which is a pretty neat experience.

And for the last project we decided together, as a class, that everybody wanted to work on apps that improve student connection on campus: to help people make friends and feel more at home at Illinois. So we will have nineteen different takes on a solution to that problem.

[![Slide 16: You say it, the agent builds it](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/16-intent.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/15)

An effective collaboration is one where the student brings their humanity: their lived experience, the things they care about in the world. The AI agent is the expert at technology. Even if AI gets tremendously powerful, I do not think it will understand human lived experience, so we will always have to explain to it what we want to happen in the human world. And because students have to put everything they want into words, I can see their entire thought process, which gives us new ways to help them learn.

### What everyone should learn

[![Slide 17: What to teach](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/17-q-what-to-teach.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/16)

A couple of takeaways before I wrap up.

[![Slide 18: Takeaways](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/18-takeaways.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/17)

Working with AI is a skill. It is not magic. Students do not all start off good at it, and they improve with practice. To help them get good at it, we should teach it early and often. AI can help with that. And students really enjoy it. It is meaningful to them, because it lets them see what they want to exist come true.

[![Slide 19: Tech without the tech companies](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/19-tech-without.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/18)

Some people think AI is going to concentrate power among the tech companies. There is another future possible here.

[![Slide 20: Few built for everyone](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/20-network-morph.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/19)

Today all of us are forced to use technology written by a small number of people.

As people start to create things on their own, we will see them building things for themselves that work exactly the way they want, or for small communities of people they know and care about. That is a much more decentralized future, about agency, not about being forced to use something a billionaire built. It is hard to care about a billion people at a time. It is much easier to care about a small group.

[![Slide 21: Four Years](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/21-four-years.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/20)

The course I am teaching is fifteen weeks, and students have already learned so much.

Imagine them doing that for four years.

It is hard for me to even conceive of that student, the capabilities they would emerge with, and how equipped they would be to use technology to make the world a better place.

[![Slide 22: What Our Students Asked For](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/22-student-asks.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/21)

I added this at the last minute. Last Tuesday, in my large introductory course, students talked with each other about AI and education, and at the end I asked them: what would you like the university to do? The number one answer was teach us how to learn with it. I find that encouraging. A lot of people think students mostly use AI to avoid learning. The students in my course want to learn. They want assignments that reward real learning, and AI that helps them think instead of handing over answers.

### Try it yourself

[![Slide 23: Try It Yourself](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/23-try-it.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/22)

If you want to try this yourself, here are a few places to start.

### Questions?

[![Slide 24: Questions](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/24-hand-off.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/23)

Thanks, everybody, for being here, and thanks to Cory for the invitation.

### Questions from the room

[![Slide 25: As AI gets more capable](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/25-qa-more-capable.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/24)

How might my classes change as AI becomes more capable? I don’t know. I feel like I am racing to catch up with these technologies myself, and what I can do in my courses has changed so much in the past year. I think it is up to us as humans to keep up, and to keep asking how we bring our humanity to these encounters. That matters more and more for higher education. We bring students from all over the world to campus, and then in some ways we do not use the fact that they are here together, in the same room, able to talk face to face. We need to do more of that.

[![Slide 26: Agents led astray](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/26-qa-instructions.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/25)

Can what an agent reads redirect its instructions or poison its memory? That is certainly possible if somebody else is able to influence your agent’s context. I have not had much trouble with it myself. There are also fears that if you let AI do certain things it might be destructive, and we have not seen that in my courses yet.

[![Slide 27: Blue books and reluctant professors](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/27-qa-blue-books.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/26)

Would I replace blue books with a bot, and how do you bring reluctant professors along? I think people will come along. We have been running meetups this semester that bring together people who are using AI in their own courses, so they can see what each other are doing and be inspired by it. Students are using AI to work around the things we want them to learn, and I think the amount of learning happening in universities has gone down, partly because we have not adjusted what we do as faculty. As for blue books: I am a strong believer in frequent assessment. In my thousand-student course there is a quiz every week, and it is hard to run a blue book exam every week. With only a midterm and a final, students do nothing, cram, and forget. The difference between cramming and studying is frequency. If you cram every week, you are no longer cramming. You are studying.

[![Slide 28: Values](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/28-qa-values.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/27)

Should a course like this also teach the values people need to use these tools well? Absolutely. I think we are going to see entire computing curricula change from top to bottom. There is a line of Stewart Brand’s that I use in these talks: we are as gods and might as well get good at it. With great power comes great responsibility, and these tools are incredibly powerful. I talked mostly about what we do with agents, but the material I am teaching in these courses is entirely new too.

### Backup slides

[![Slide 29: Backup Slides](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/29-backup-slides.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/28)

These are slides I did not get to in the talk.

[![Slide 30: One student, two agents, a room](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/30-agent-buildup.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/29)

In the no-code course every student works with an agent that builds.

A second agent watches the conversation.

At every table.

And across the room, so the instructor knows who to talk to, and about what.

[![Slide 31: What did you see?](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/31-step-in.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/30)

Here is the second agent stepping in. The builder says it fixed the problem.

Then it asks the student: what did you see?

The window still disappeared.

So it fixes that first. The watcher gives the builder context. It never instructs the student.

[![Slide 32: Down from the mountain](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/32-mount-agent.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/31)

What we teach students about talking to an agent: say what you want. Discuss ideas, not implementation.

[![Slide 33: A lesson inside their own agent](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/33-agent-lessons.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/32)

In the large course, a lesson is a page with a prompt on it.

The student pastes it into their own agent.

The lesson then runs inside their own project, on their own prompts.

[![Slide 34: Using and Understanding AI: Final Projects](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/browse/34-showcase-uai.jpg)](https://www.geoffreychallen.com/talks/2026-11-20-pitt-osher/slides.html#/33)

And the final projects from Using and Understanding AI.
