This past summer (July and August here in Calgary), I taught EDER 619.26, Leadership for Learning: Policy, Governance, and Community, a fully online course in the University of Calgary’s Master of Education (MEd) program. Most of my students were K-12 teachers in Alberta pursuing advanced training while working full time. This was my first time teaching the course, and I used it as an opportunity to redesign my assessment approach in response to a concern I have followed closely in recent months: the threat that agentic AI poses to academic integrity in online courses.
I want to share what I did, why I did it, and how another instructor could try the same approach.
The Problem I Set Out to Solve
In online courses we have relied on the same assessment tools for decades: discussion board posts and a final paper submitted at the end of the term. These formats work well when the greatest risk to academic integrity is a student copying a classmate’s work. They work less well when a student can hand an assignment to an AI agent and receive a finished product in return. I wanted an assessment structure that valued process over product and made real-time, verifiable engagement central to how students earned their grades.
My Approach
I built the course as a hybrid model with mandatory video conference (i.e., Zoom) sessions, scheduled well in advance so students knew what to expect. I selected a small number of required readings and asked students to locate supplementary readings on their own through the library databases, connected to the weekly themes and the in-class tasks. I adopted a flipped classroom structure: students completed the readings before each session and arrived prepared to apply them.
Seventy-five percent of the course grade came from three real-time learning tasks tied to the Zoom sessions. For two of the three sessions, I brought a current Canadian policy or governance news story to the group, chosen because the course readings focused on the Canadian context. For the third, I built a synthetic educational case with the assistance of Claude, and I told my students I had done so. In each session, students worked in self-selected groups for a set period, initially ten minutes and later shortened to five, to apply the readings to the case or news story and produce a shared artifact documenting their thinking. They then uploaded that artifact to a course Dropbox.
I informed students from the outset that I was trying this format for the first time and that I expected to adjust it as I learned what worked. I asked for their feedback throughout the term and changed the process in response, including the shortened submission window after students reported that some groups kept working past the agreed time.
I also changed how I graded. I was not looking for a polished, consensus-driven product. I wanted evidence of student thinking, including open questions and points of disagreement that a group had not resolved. I told students that AI tool use was permitted under the University of Calgary Faculty of Graduate Studies Artificial Intelligence Guidelines and would not affect their grade either way. Most reported that they spent the bulk of their time in conversation with their groupmates rather than using AI tools, largely because the time constraint left little room for anything else.
The remaining twenty-five percent of the grade was a synthesis paper. Students combined the required and supplementary readings, the bibliographic sources they had shared with classmates during the sessions, and the two news stories and the synthetic case addressed in class into a single integrated paper.
Steps to Try This Assessment Approach
- Schedule your synchronous sessions early. Set the dates and times for all mandatory sessions at the start of the term and communicate the expectation of real-time participation.
- Curate a limited reading list. Choose a small set of required readings and ask students to locate supplementary sources through the library databases, tied to the weekly themes.
- Flip the classroom. Assign readings for completion before each session and communicate this expectation in writing.
- Prepare a case or current news story for each session. Select material relevant to your discipline and your students’ context, or construct a synthetic case with AI assistance. If you use AI assistance, disclose it to your students.
- Set a group task with a firm time limit. Give students 5 to 10 minutes to work in self-selected groups, apply the readings to the case, and produce a shared artifact documenting their discussion and reasoning.
- Collect the artifact through a shared drop point. I used the D2L / Brightspace Dropbox, but you could also use a shared document, or similar tool works well. Set the submission window based on your own testing; shorten it if students report that groups continue working past the agreed time.
- Grade for process, not polish. Communicate to students that you are evaluating evidence of learning and engagement with the readings, not a finished, consensus-driven product. Tell them unresolved questions and disagreements are acceptable and worth documenting.
- State your AI expectations explicitly. Clarify whether AI tool use is permitted for the in-class task and confirm that it will not affect grading either way, consistent with your institution’s guidelines (if they have them).
- Build a synthesis assignment. Ask students to integrate the required readings, their self-sourced supplementary readings, and the material from each session into a single paper at the end of the term.
- Offer an alternate path for students who miss a session. Schedule a makeup session or provide an equivalent assessment based on the same weekly reading and format.
What I Would Change Next Time
Students told me they would have benefited from more time to discuss each case or news story during the sessions. I plan to extend the collaboration window in future offerings of the course.
Reflection
Most of my students had not encountered agentic AI before this course and some did not know the term. That gap became a useful discussion point, even though artificial intelligence and academic integrity were not the stated focus of the course. Students reported that the sessions felt more purposeful than a standard discussion board, in part because they knew each session would produce a graded outcome. I plan to repeat this assessment structure. This updated assessment approach replaced a format that has gone unchanged in online learning for decades and shifted the emphasis toward collaboration, real-time problem solving, and process over product.
This was by no means a perfect experiment, but in the end, the effort was worth it… And I’ll close by saying that the students were — and are — brilliant, thoughtful, and inspiring.
___________
Share this post – Redesigning Assessment for the Age of Agentic AI: A Real-World Reflection – https://drsaraheaton.com/2026/09/23/redesigning-assessment-for-the-age-of-agentic-ai-a-real-world-reflection/
Sarah Elaine Eaton, PhD, is a Professor and Research Chair in the Werklund School of Education at the University of Calgary, Canada. Opinions are my own and do not represent those of my employer.
Posted by Sarah Elaine Eaton, Ph.D. 
You must be logged in to post a comment.