What I Learned About Assessing Students in the Age of AI

Before developing my arguments, let me share two recent anecdotes from an AI introduction class I taught in 2026 to a group of business students.

The first occurred during a 40-minute group exercise. The objective was to find 10 use cases for AI in the financial sector, prioritize them, select the project with the highest potential, and detail a transition plan from business case to project plan. The output was to be presented via slides, and students were explicitly allowed to use AI.

After just five minutes, a student raised his hand and asked what they should do with the remaining 35 minutes. When I designed the exercise, I assumed the 40-minute time constraint would be tight. I never anticipated students would skip the process entirely, prompt an AI for a full slide deck, and consider the job done.

Caught off guard, I addressed the room: "I assume most of you already have your final presentations ready?" Nods spread across the room. "Alright. You now have 35 minutes to actually read and understand what is on those slides, because I will be questioning you on every detail."

When the presentations began, the content was visually polished and logically structured, but the students were merely reading text they had never processed. I faced an immediate pedagogical dilemma: do I let them present as planned or do I step in so they can understand better? I chose to intervene, turning their presentations into real-time teaching sessions to ensure they walked away with at least a basic understanding of the concepts.

The second anecdote took place during a final project presentation. Students had two weeks to build a chatbot that could process customer orders for a pizzeria. They were required to present the concept, brand name, menu, and functionalities, execute a live demo, and reflect on their lessons learned. I provided generic chatbot code bases to start from and told them they were welcome to contact me the entire time if they had doubts or needed help building it.

Throughout the two weeks, not a single student reached out. I grew suspicious: Had they all quietly mastered chatbots when it took me months?

On presentation day, the live demos ran flawlessly. Some students used Google Gems, while others leveraged Gemini or Claude with Canvas features enabled to display interactive mock phone interfaces. When I asked one student how long his team spent on the assignment, he proudly replied: "15 minutes."

Once again, the students were presenting material they hadn't truly understood. Even the "lessons learned" section was entirely AI-generated. I found myself stepping in again, explaining why certain architecture choices were superior to others and transforming their presentations into hybrid lectures.

In both instances, the outcomes missed the mark. It wasn't a total loss—students were fascinated by what AI enabled them to generate, and they did learn something. But they learned far less than intended. Furthermore, genuine collaboration vanished. In the past, students would divide tasks, debate approaches, and problem-solve together. With AI doing the heavy lifting, peer interaction dropped to zero.

When I asked the students why they relied on AI for every single step, their response was pragmatic: "We are overwhelmed with assignments across all our courses. We need AI to meet deadlines, and since everyone else is using it, we’d be at a disadvantage if we didn’t." Underneath that real pressure lies a natural human instinct to seek the path of least resistance.

Assessing them was equally complicated. In the past, I would have graded them on the quality of the end result and the presentation, but both were now generated by AI. So, I adapted and decided to grade them on how well they seemed to understand what they were presenting. By asking a few targeted questions, you can easily evaluate their true level of understanding.

Reflecting on these experiences, I have categorized how our approach to student assessment must evolve into four core pillars.

1. Reframing Evaluation & Grading Criteria

We can no longer assume that a polished deliverable implies subject mastery. Assessment structures must explicitly decouple execution from comprehension:

• Split the Grading Matrix: Divide marks clearly between AI tool execution and subject mastery. For example, assign 30% to the final deliverable, 30% to oral presentation delivery, and 40% to a live Q&A defense.
• Encourage quality particiaption: A student’s curiosity, engagement, and consistent effort should carry significant weight—potentially up to 50% of the overall grade. To keep this objective and defendable, ground it in structured oral defense criteria and systematically track insightful contributions and questions during class sessions.
• Rely on Rigorous Q&A: Every presentation must be paired with an extensive Q&A session designed to test deep understanding. Students should know in advance that they will be expected to explain and defend every claim, visual, or line of reasoning in their deliverable.

2. Redesigning Assignments for Active Learning

When AI can generate a finished product in seconds, assignments must focus on the process of learning rather than just the final results:

• Give instructions step by step: Prevent students from jumping directly to the final output by releasing instructions step-by-step. Forcing students to navigate intermediate phases prevents the "5-minute generation" shortcut and ensures they engage with each stage of the problem.
• Reintroduce Low-Tech and Offline Environments: While restricting AI access entirely is difficult, controlled offline activities remain crucial. Incorporate pen-and-paper exercises, physical card games, or live classroom dynamics where students must think on their feet without digital assistance.

3. Cultivating Classroom Culture and Human Skills

Technological shortcuts risk eroding the psychological benefits of intellectual struggle and peer collaboration:

• Foster a Culture of Effort: True learning satisfaction stems from overcoming genuine obstacles. Are you happier when you get a good grade by coasting through an exam, or when you study hard and earn top marks? We must remind students that struggling through a concept builds mental resilience and long-term retention. The brain is a muscle and exercising it yields long-term benefits.
• Engineer Real Peer Collaboration: Because AI eliminates the need to divide labor or ask peers for help, students are becoming increasingly isolated. We must design group activities that strictly require human interaction and hold individuals accountable by for example asking team members at random to justify specific design choices made by the group.
• Incorporate Interactive Classroom Dynamics: Shift class time away from static presentations and toward active intellectual formats like debates, structured discussions, and live problem-solving, where genuine comprehension and critical thinking are immediately visible.

4. Integrating AI Literacy and Critical Auditing

Since banning AI is neither realistic nor desirable, our goal must be to transform passive copy-pasting into active critical evaluation:

• Teach Critical Auditing: Copying AI outputs verbatim reflects uncritical trust. Students must be taught to verify sources, cross-reference outputs with alternative tools, and challenge underlying assumptions.
• Design "Fact-Checking" Exercises: Create assignments specifically around auditing AI work. For instance, provide students with an AI-generated analysis containing subtle flaws, hallucinated data, or weak logic, and grade them on their ability to identify, correct, and prove those errors.

Final Thoughts

While I remain fascinated by what modern AI can achieve, its unguided integration into education presents a clear risk. Understanding in retrospect after an AI generates a solution is never as effective as learning forward through trial, error, and problem-solving. AI does not just retrieve information; it bypasses the precise cognitive friction required to learn.

As educators, our role is not to block these tools, but to redesign our teaching environments so that students use AI responsibly—ensuring that the essential struggle of learning is preserved while preparing them for the reality of tomorrow.