How AI Teaching Assistants Are Transforming Online Graduate Programs
Professor Anna Marbut on building and deploying OpenAI-powered TAs in asynchronous, professional-focused courses
Anna Marbut teaches Applied Artificial Intelligence and Applied Data Science at University of San Diego.
Source note: This is an edited interview adapted from a narrated video submitted to OpenAI. Watch the associated video in OpenAI Academy.
Intro
What happens when you bring AI-powered teaching assistants into a fully online, asynchronous graduate program? Anna Marbut, Professor of Practice at the University of San Diego, has spent the past two years answering that question. In this conversation, she shares the story behind her team’s integration of OpenAI’s Assistant function into their Applied Artificial Intelligence and Data Science masters programs, how the system works, what the impact has been, and where she sees the greatest potential for students and faculty alike.
The Interview
Q: Anna, let’s start with the basics. What motivated you and your team to bring AI teaching assistants into your online programs?
Marbut: Our programs are fully online and asynchronous, serving mostly working professionals—people advancing in their careers or changing industries. About two years ago, we realized that the lack of real-time interaction was a challenge. Students needed more immediate support, especially outside of traditional office hours. That’s what led us to experiment with AI teaching assistants, accessible through our Slack channels.
Q: How did you go about building these AI TAs? What does the setup look like behind the scenes?
Marbut: We used OpenAI’s Assistant function. In our OpenAI API account, we set up assistants for each course, plus one for administrative questions—things like enrollment, payments, and deadlines. Each assistant is configured with system instructions: what the course covers, how to respond, and, importantly, what not to do. For example, we tell it not to give out direct answers to quizzes or assignments. We also uploaded a full dump of our Canvas course materials—discussion prompts, quizzes, assignments—so the assistant can use retrieval-augmented generation (RAG) to answer questions based on actual course content. Since our students often have coding questions, we enable the code interpreter as well.
Q: How do students actually interact with the AI assistants?
Marbut: We integrated the assistants into Slack using a third-party connector. Students can message the TA privately or tag it in any course channel. For example, if a student is stuck on a homework assignment, they can ask for tips and get a response directly in the thread. The assistant references the relevant course material and provides guidance—sometimes even example code—but always avoids giving away answers outright. If a student tries to get the quiz answers, the assistant politely refuses and offers study tips instead.
Q: What kind of impact have you seen so far—on students, faculty, and administrators?
Marbut: Since launching, our AI TAs have answered about 200 questions in eight months. That’s 200 questions our professors and administrators didn’t have to handle directly. Faculty have noticed a shift: the questions that come up in office hours or on Slack are deeper and more thoughtful—students seem to be doing more troubleshooting on their own before reaching out. Administrative staff are also getting fewer routine questions about enrollment or deadlines, since the bot handles those. Of course, we’d like to see even more usage, so we’re working on messaging and fine-tuning the bots based on student feedback.
Q: What are some of the challenges or next steps?
Marbut: The biggest challenge is awareness—making sure students know these tools are available and feel comfortable using them. We’re also continually improving the assistants, both in terms of accuracy and helpfulness. In a fully online, asynchronous program, having a 24/7 TA is a game-changer, especially for working professionals who study after hours or on weekends. But we’re still learning how to maximize their impact.
What Stands Out
Core idea: AI teaching assistants provide immediate, 24/7 support in fully online, asynchronous graduate programs serving working professionals.
Classroom design: The AI TAs are integrated with Slack and Canvas, configured to guide students using course materials without giving direct answers to assessments.
Student impact: Students receive timely help that encourages independent troubleshooting, while faculty handle fewer routine questions and focus on deeper interactions.
Transferable lesson: Thoughtful integration and clear boundaries for AI assistants, combined with ongoing feedback and communication, enhance support in online education.
Bio
Anna Marbut is a Professor of Practice in the Master’s in Applied Artificial Intelligence and Master’s in Applied Data Science programs at the University of San Diego.