The final product is dead as evidence of learning

For decades, the university has assessed learning through products: essays, reports, presentations, code. The premise was simple: the product was a reasonable proxy for the process that generated it.

Generative AI has broken that premise. The proxy no longer represents the process.

What AI cannot simulate in a sustained way

The answer is not to ban, nor to detect, nor to return to the paper exam. It is to redesign what we observe:

• Reasoning: not the answer, but the path.
• Decisions: what was discarded and why.
• Iterations: how the work evolves with feedback.
• Transfer: applying what was learned in a new context.
• Oral defence: sustaining knowledge without intermediaries.
• Metacognitive reflection: explaining what one has learned about oneself.
• Quality of AI use: as a competency, not as cheating.

From detection to design

AI detection tools fight a losing battle. Authentic assessment design wins the war without fighting it: when the evidence lies in the process, the question “did the AI do it?” loses relevance, because AI is a legitimate part of the process and learning is demonstrated elsewhere.

This transition — from product to process — is probably the greatest pedagogical redesign universities will face this decade. And it is an institutional architecture decision, not the work of one brave professor.