Over the past few weeks, I finished teaching Consulting in Hospitality & Tourism at Nova SBE.
On paper, it is a relatively simple course: around 40–55 master's students every year, divided into consulting teams, working on real-world tourism projects over six intensive classes.
But this year, the course became something else entirely.
Not because of the content.
Not because of the projects.
And certainly not because AI suddenly replaced professors.
What changed was the operating system surrounding the learning experience.
This course became part of a broader experiment we are exploring at Nova SBE within the discussion around the "Future of Universities": how should universities actually operate in a world where Artificial Intelligence is fundamentally changing knowledge work?
Not theoretically.
Operationally.
The Structure of the Course
Consulting in Hospitality & Tourism is intentionally practical and sprint-like.
The students are all master's students, mostly from Nova SBE's Master's in Management, consistently ranked among the best in the world. This matters because these are students who will soon enter the job market in consulting, strategy, hospitality, tourism, technology, finance, and many other sectors where AI will not be optional. It will be embedded in the way work gets done.
This year, we had a little over 40 students, divided into six groups.
The groups are intentionally large. In a trimester course with weekly deliverables and a demanding workload, this allows students to distribute responsibilities across the team. But it also creates a very real management challenge: coordination, accountability, ownership, and contribution quality all become part of the learning process. That is why peer assessment remains an important part of the course design.
Each class corresponds to a consulting project deliverable:
- Project Plan
- As-Is / Problem Definition
- Benchmarking & Foresight
- Gap Analysis
- Recommendations
- Final Presentation
Every week, students present their work to the professor, and in the final stage, always with the client in the room.
Over the years, this course has worked on real projects with municipalities such as Torres Vedras and Mafra, with Turismo de Portugal, and with the Portuguese Secretary of State for Tourism.
This year, all projects were developed for Vila Galé — one of the top hotel chains in Portugal.
So the course is not designed as a traditional academic exercise. It is designed as a consulting sprint, with real clients, real ambiguity, real deadlines, and real consequences.
And that distinction matters.
The Traditional Model Was Starting to Break
Historically, the course followed a fairly conventional university setup:
One professor.
One Teaching Assistant.
Weekly classes.
Optional office hours.
The problem is that this model increasingly struggles under modern expectations.
Students expect faster feedback. Projects evolve weekly. Iteration cycles matter. And office hours almost never worked properly.
Either everyone showed up at once, creating chaos, or nobody appeared at all.
At the same time, traditional written deliverables were becoming harder to interpret as genuine evidence of learning in a world where AI can generate polished outputs extremely quickly.
So this year, we redesigned the course around a simple assumption:
AI already exists.
Students are already using it.
So pretending otherwise is probably the least productive option.
1. The Teaching Assistant Was Replaced by AI Support
This does not mean AI replaced human mentorship.
It means AI replaced part of the operational support layer surrounding the course.
Instead of relying on a traditional TA structure, AI became a force multiplier for the professor.
It supported:
- generating detailed feedback;
- synthesizing observations;
- identifying logical weaknesses;
- structuring recommendations;
- comparing the evolution of each group across deliverables;
- and accelerating iteration cycles between classes.
The result was surprisingly important.
Students received significantly more feedback than before, in less time, and with more consistency.
More importantly, it allowed the professor to spend less time on operational repetition and more time on actual coaching, judgment, and challenge.
Ironically, AI made the course feel more human, not less.
2. Feedback Became Continuous Instead of Episodic
During class, groups presented their weekly deliverables live and received immediate feedback in the room. Then, within roughly 48–72 hours, each group received detailed written feedback generated with AI support and refined by the professor.
This changed the assessment logic. The course became less dependent on evaluating a static written document and more focused on:
- quality of reasoning;
- clarity of communication;
- live discussion;
- defense of ideas;
- iteration capability;
- presentation quality;
- and team dynamics.
In other words: less emphasis on the artifact, more emphasis on the thinking process behind it.
And that distinction becomes increasingly important in the AI era.
When a written deliverable can be polished by AI, the classroom must become a place where students are asked to explain, defend, adapt, and own their thinking.
3. Students Were Encouraged to Use AI — But Required to Disclose It
One of the biggest decisions was philosophical.
We decided not to fight AI usage.
Instead, students were actively encouraged to use AI throughout the consulting process, provided they disclosed how they used it.
This immediately changed the classroom dynamic.
The discussion stopped being: "Did you cheat?" And became: "Did you think?"
Some groups used AI exceptionally well. They used it to accelerate research, stress-test ideas, improve structure, simulate client perspectives, benchmark international examples, prepare interview guides, improve slides, and iterate faster.
Others over-relied on it. And the consequences became visible very quickly. Some students produced polished intermediate deliverables that looked impressive on paper but later struggled during live presentations. They read directly from slides, lacked ownership of the reasoning, and could not properly defend the logic behind the recommendations.
That was probably one of the most important learnings of the entire experiment. AI can amplify thinking. But it can also amplify intellectual passivity. And classrooms are one of the few places where that difference becomes immediately visible.
4. Prompt Engineering Became Foundational Literacy
In the first class, students received training on AI fundamentals, prompt engineering, and effective AI collaboration workflows.
Not because prompt engineering is a trendy skill. But because structured prompting is increasingly becoming a form of structured thinking. Students who provided better context, framed better questions, and iterated more intelligently consistently produced stronger outputs. The quality gap was rarely about the model itself. It was about judgment.
Students who used AI well did not simply ask for answers. They used it as a thinking partner. They questioned it, challenged it, refined it, and combined it with their own analysis.
That is a very different skill from simply generating content. And it is one universities will need to teach explicitly.
5. AI Made the Flipped Classroom Operationally Viable
Traditionally, the first class included a large theoretical block on structured communication and the Minto Pyramid Principle.
This year, that component became pre-work.
Students studied the material asynchronously before class through AI-supported learning artifacts created in Claude.
That changed the first class significantly.
Instead of spending too much time transmitting concepts, we used classroom time for discussion, application, coaching, and debate.
For years, universities have discussed flipped classrooms.
AI is the first thing I have seen that makes them operationally scalable without dramatically increasing preparation complexity. It allows professors to create better pre-work, more interactive study materials, tailored explanations, quizzes, examples, summaries, and learning artifacts faster than before.
But the real value is not the artifact itself.
The real value is that classroom time becomes more precious. Less transmission. More application. Less passive listening. More active learning.
6. WhatsApp Worked Better Than Formal Academic Communication Tools
Although Teams and email remained available, the primary communication channel became WhatsApp between the professor and each project group.
This was probably one of the least academic — and most effective — decisions. The communication became faster, more contextual, less bureaucratic, and significantly more natural. In a trimester course operating like a weekly sprint, eliminating friction mattered enormously. Small questions were solved immediately. Momentum stayed high. Students felt accompanied without needing formal office hours. And surprisingly, boundaries became clearer rather than blurrier.
This does not mean WhatsApp should replace every institutional platform. But it does suggest that the design of communication channels is part of the learning model. If the course operates like a sprint, communication also needs to operate like a sprint.
What We Learned
1. AI Works Best as a Leverage Layer for Professors, Not as a Replacement
The future is probably not AI versus professors. It is professors amplified by AI versus educational systems designed for a pre-AI world. The value of human judgment, mentorship, energy, and real-world experience actually increases once operational friction decreases. AI did not make the professor less relevant. It made the professor more available for the parts of teaching that matter most.
2. Universities Need to Rethink What "Proof of Learning" Means
If students can generate sophisticated documents in minutes, then static deliverables become insufficient as standalone proof of competence. The ability to explain, defend, adapt, synthesize, improvise, and communicate live becomes far more important.
This does not mean written work disappears. But it does mean that written work cannot be the only evidence of learning. The process matters more than the artifact.
3. AI Exposes Weak Thinking Very Quickly
One unexpected insight was this: AI does not eliminate rigor. It exposes where rigor was fake. Students who deeply understood their projects became dramatically stronger with AI.
Students who outsourced thinking became easier to identify, not harder. In many ways, AI did not create the problem of shallow thinking. It simply made it more visible.
4. Speed Matters More Than Universities Realize
Modern knowledge work operates through rapid iteration loops. Weekly feedback matters. Fast responses matter. Continuous adjustment matters. AI compresses learning cycles dramatically. Universities that continue operating with slow feedback systems and highly bureaucratic interaction models may increasingly feel disconnected from how real-world knowledge work actually functions.
5. Students Want Autonomy, But They Also Need to Feel Accompanied
One of the strongest signals from this experience was that many students felt both more autonomous and more supported.
That combination is powerful. They were not being micromanaged. They had freedom to use tools, divide work, explore directions, and make decisions. But they also had faster access to feedback, clearer next steps, and more frequent guidance. This may be one of the most important design principles for the future of education: More autonomy does not mean less support. It means better support, delivered at the right moments.
What Other Universities Can Take From This
This was not a technology experiment. It was a course design experiment. And that distinction matters because the main lessons are not dependent on one specific AI tool, platform, or model. They are about how universities can redesign learning experiences when AI becomes part of the normal workflow of students. Here are a few principles I believe other universities can take from this experience.
1. Start With the Course Operating Model, Not the Technology
The first question should not be: "What AI tool should we use?" The better question is: "Where is the current learning experience too slow, too rigid, or not creating enough value for students?" In this course, the bottlenecks were clear: feedback cycles, office hours, written deliverables, and limited professor bandwidth. AI became useful because it addressed those bottlenecks.
2. Treat AI as Infrastructure for Better Teaching
AI should not be positioned only as a student productivity tool. It can also become teaching infrastructure. It can help professors prepare materials, generate feedback, create simulations, build prework artifacts, synthesize student submissions, and identify patterns across groups. The objective is not to reduce the role of the professor. The objective is to increase the professor's leverage.
3. Move Assessment Closer to Live Reasoning
If AI can help students produce polished written documents, then universities need to put more weight on what AI cannot fully replace: live explanation, discussion, judgment, critical thinking, and the ability to defend recommendations. This does not mean written work disappears. But it does mean that presentations, oral defense, iteration, and in-class discussion become more important evidence of learning.
4. Make AI Usage Explicit, Not Hidden
Trying to ban AI often creates an artificial classroom reality. Students will use it anyway. A more productive approach is to require disclosure. How was AI used? For research? For synthesis? For structure? For writing? For slide improvement? For idea generation? This moves the conversation from policing to learning. The key question becomes not whether students used AI, but whether they used it with judgment.
5. Redesign Feedback Loops
In many university courses, students receive feedback too late for it to meaningfully improve their work. AI allows feedback cycles to become faster, more detailed, and more frequent. That can fundamentally change the learning experience. The value is not simply that feedback becomes quicker. The value is that students can act on it while the project is still alive.
6. Use AI to Make Flipped Classrooms Easier to Execute
Many professors like the idea of flipped classrooms, but they are difficult to operationalize at scale. AI can help create pre-work materials, interactive study guides, summaries, quizzes, and learning artifacts that allow students to prepare before class. That makes classroom time more valuable. Instead of using class to transmit information, universities can use it for discussion, coaching, application, and judgment.
7. Do Not Underestimate Communication Design
One of the most effective changes in this course had nothing to do with advanced AI. It was using WhatsApp as the main communication channel with groups. Universities often focus on formal platforms, but students operate in faster, more fluid communication environments. If the learning experience is sprint-like, communication also needs to be sprint-like. The design of communication channels is part of the pedagogy.
8. Accept That Some Students Will Misuse AI — And Design for That
Some students will over-rely on AI. That is not a reason to ban it. It is a reason to design courses where over-reliance becomes visible. Live presentations, client discussions, oral defense, peer assessment, and iterative feedback make it much harder for students to hide behind AI-generated work. In that sense, well-designed AI-enabled courses may actually become more rigorous, not less.
9. Build Small Experiments Before Trying to Redesign the Whole Institution
Universities do not need to transform everything at once. They need more controlled experiments inside real courses. One course. One cohort. One assessment model. One feedback loop. One redesigned office hours system. The future university will not be discovered through strategy documents alone. It will be discovered through experiments in actual classrooms.
Final Reflection
At Nova SBE, we have been encouraging the responsible use of AI since 2023. Since late 2025, we have also provided students with access to AI tools, making sure that the discussion is not only conceptual, but practical and embedded in their daily learning experience.
Now, the next challenge is different. It is no longer just about giving students access to AI. It is about understanding how AI should reshape the operating system of the university itself. How should courses be designed? How should feedback work? How should students be assessed? What should happen inside the classroom? What should happen before and after class? What is the role of the professor? What kind of skills should students develop before entering the job market? This course was one small experiment inside that broader transformation. It did not provide all the answers. But it reinforced one conviction: The future of universities will not be built by adding AI on top of old models. It will be built by redesigning the learning experience around a new reality — one where AI is already part of how students think, work, research, create, communicate, and learn.
Continue to Part 2 → A deep dive into how the students actually used AI — in their own words.Originally shared on my Substack →