Dr. Shane Saunderson is an Assistant Professor of Information Systems with the Degroote School of Business at McMaster University and the Director of the Research on Artificial Persons (RAP) Lab. His research focuses on the social and organizational implications of humanlike technologies, such as robotics and AI, with particular interest in topics such as trust, persuasion, and anthropomorphism. He teaches Commerce 3AI3 – Foundations of LLMs and GenAI in Business – a synchronous, in-person course in the new McClean Center for Collaborative Discovery, where he explores experimental pedagogy that leverages AI as both a topic and a tool in the classroom. For a more detailed overview of Dr Saunderson’s work, see his full biography.
Compared to what students will encounter in the real-world, a written case has a major structural weakness: it lacks interaction. Everything a student needs arrives on the page at once, evenly lit and equally available. Nothing has to be asked for, probed, challenged, or explored. In reality, analysts spend most of their effort deciding whom to talk to and what to ask them; case students spend most of their time deciding what to summarise. GenAI sharpened that point; human value is no longer in a well organized, written deliverable, but in someone’s ability to investigate and uncover insight.
Shane’s response to the existing case study was to build the part that had been missing: interaction.
The result is an interactive case study that redefines case-based learning. The student is cast as an external strategy consultant, retained by a business to answer a key strategic question about the future of that business. A short brief supplies high-level, public data and nothing else; the meat of the case is not on the page. To get it, they have to interview the “people” who have it.
Four executives sit behind four cards: a chief executive officer, a chief financial officer, a chief technology officer, and a chief marketing officer (Figure 1). Each is a separate GenAI agent with its own briefing, its own temperament, and its own slice of the facts. The slicing is deliberate: the CFO holds the detailed financials, the CTO knows what could be built and what it would cost, the CMO knows what customers will accept, and the CEO carries the strategic vision. They answer what is asked, not what was meant, and none of them will do the work for you: push one for a recommendation and it declines with the tone of someone paying for advice rather than giving it.
Some agents are holding a decisive insight that only a well‑aimed question will surface, and this is where the teaching happens. Ask the CEO what the strategy is and they describe the current plan. Ask what comes next and they describe the unannounced plan, which reframes the entire case. Ask the CFO whether a business unit is profitable and the answer could be read in a financial statement; ask what offerings are growing and the number underneath it comes out. These insights sit one good question away, but require students to learn how to ask the right things of the right people. When an insight surfaces, the page registers it and tells the student a key finding has come out, so the unlock is legible rather than silent.
Access is scarce. The CEO offers roughly five substantive questions before ending the interview for another meeting. A clock advances in the corner of the conversation, they grow visibly more pressed, and once they leave the interview closes for good. Most students will spend two of their five questions on things the CFO could have told them, and terse replies from the CEO become part of that lesson.
When the interview phase is complete, the student writes a recommendation and takes it into a virtual boardroom (Figure 2), where all four executives read and challenge it in turn. Thin answers get probed and challenged. Well thought-out answers get built upon and supported. Claims that stray into someone else’s territory get intercepted: promise a two‑quarter build in front of the CTO and they will cut in with heavy skepticism.
A marked debrief follows: what each agent judged to have been addressed against a named set of concepts, a short verdict in each of their voices, any insight left uncovered in the room, and a conclusion to the case of what the company actually did in the end. Beneath the case sits the full transcript of every question asked.
Two system constraints are critical for making the interactive case work. The agents have no internet access and no knowledge of anything after the date the case is set, because a well‑connected model would be able to tell a student how the story ends. In addition, student messages are treated as dialogue, never instruction, so attempts to talk an executive out of its role are answered in character rather than obeyed. Building the page took an afternoon with Claude Code; developing a plausible scenario and the four executives took considerably longer.
The simulation has not yet met a class. The pilot is still being planned, so what follows is intent rather than evidence.
The goal here is to augment the traditional case with a new set of learning objectives. Students have to get something out of another party: decide who is worth their time, open with a question that earns a real answer, and recognize when they have been handed a frame rather than a fact. We are no longer training analysts because analysis is the easy task we can give to GenAI.
Recommendations must be built on partial and inconsistent testimony. The four executives do not always agree, their accounts are incomplete by design, and nothing arrives pre-sorted. The student has to reconcile four versions of the same case and judge which numbers and responses actually matter, rather than summarising a page that has already done the sorting for them. We are training strategists who can live comfortably amidst ambiguity, not cogs in a machine that can churn out widgets through a framework.
Their recommendation is then only worth what a student can defend. The boardroom tests whether a position survives contact with the people it affects: whether a student can hold a line under pressure, concede the point that should be conceded, and tell the two apart. The hope is that this design affords them space to reflect on their own recommendations, become more comfortable with incomplete information, and embrace being challenged and disagreed with. Their efforts can then be synthesized and reviewed by peers and/or instructors to give feedback not only on their interviews and recommendations, but on how they defended and evolved their position. We are trying to training leaders for an increasingly chaotic world that can make hard calls under partial information and evolve their own thinking when the situation demands it.
The particular case developed is less important than the pattern. The mechanics are written separately from the case content, so the same structure can be aimed elsewhere: describe the organization, name the people, decide what each is holding back, and the simulation assembles around it. A colleague’s case in operations, health policy, or public administration could work the same way.
The next step is to take this case off the page and into a room. A live session is being planned at the KPMG Ignition Lab in the new McLean Centre for Collaborative Discovery building this academic year. Participants will interact with the same four executives in parallel and, after finishing the case, will spend time reflecting and discussing with their peers about what each surfaced. That comparison shows the true value of the case: the same hour and the same four executives yield a room full of materially different transcripts, and what one participant reached in three questions another never found at all. Seeing this outcome side by side makes the lesson land. The case was not the hard part; the questions were.
—
Saunderson, S. (2026), The Case Study That Talks Back: GenAI Agents Running a Live Case, Teaching in the Age of AI: Examples. Retrieved from The Case Study That Talks Back: GenAI Agents Running a Live Case, Licensed under Creative Commons BY-NC-SA 4.0.
Teaching in the Age of AI