The situation
A university master's programme in corporate communication had to work a case with AI for a day, inside the real platform a company works in, with internal data. The requirements pulled in two directions. Students had to get in within seconds, without signing up. The company couldn't lower its guard on its own platform to let a class in.
What I did
We prepared the day with Andrea Zennaro. I built the workshop mode, gave the demo and followed the groups remotely. You got in with a QR code projected in the room, your first name, last name and a personal code to come back in. Behind the QR code every student had their own identity, with permissions limited to the day and every platform control still on.
I made two other decisions. Each group had a public summary of its direction, and the AI suggested where two teams could help each other using only those summaries, without showing anyone's work. And the final vote followed the criteria in the course brief, with a mandatory explanation that every group received anonymously.
What changed
Andrea's verdict was “The comments were very enthusiastic”, with two specific reservations. Getting in and back in was too laborious. And the assistant used vocabulary that was too advanced for people outside the trade. I've already rewritten the instructions for short, simple answers, and entry is the first thing I'm simplifying. The format can be reused with other classes.
What I learned
Students got in with a QR code projected in the room, their full name and a personal code to come back in. For a single day, a shared login per group or a guest pass looked proportionate, and would have lost who wrote what, per-person limits and the ability to shut out one student. So behind the QR code everyone had a real identity. The cost showed at entry, the slowest part: Andrea found getting in and back in too laborious. He and I had tested it on a laptop and a phone, and we both know the platform. The next rehearsal is with people who have never seen it.
In the first idea for the day, each group could read the others' work to find where they could help each other. Collaboration seems to mean opening the documents. To coordinate, though, a group only needed to know the others' direction, what they were missing and what they could offer, and reading everything would have exposed work in progress to teams competing for the same vote. I changed it before the event: each group had a public summary of its direction, and the AI suggested connections between teams using only those summaries. Today, when different groups share a space, I first decide what they need to see to collaborate.
The first version of the vote asked for your favourite group. It was the easiest to build and felt right for a workshop. Next to the course brief, it measured something else: the brief called for an assessment by criteria, and with only a few groups a preference vote leans heavily on group size, without telling anyone why the winner won. The vote became a scorecard with a score per criterion and a mandatory explanation, which every group received anonymously. Now, before the day, I check the tool against the brief line by line.
In the sessions I had followed at Bocconi, students had to create an original, feasible marketing concept and explain its value on a single slide, with no working demo. The sequence was fixed: intent first, then many rough concepts, then the stress test of who wants it and why now. With AI you reach something presentable so fast that the exercise can end before anyone asks whether the idea deserves to exist, and a vote on the tools is won by whoever already knew them. That's why in the workshops I design the demo is optional and the idea is what gets judged.
The prototype isn't the destination. The loop is.