AI literacy is often taught as a set of rules: what AI can and can’t do, how to prompt it responsibly. This session works from a different premise: that students build a far stronger, more durable understanding of generative AI, and a genuine sense of ownership over it, by directing what it makes than by only being told what it does. When learners steer the process instead of just receiving its output, they stop being users of AI and start becoming its architects.

That premise extends one step further: teachers can only pass on agency they’ve exercised themselves. This session is built for teachers first — not because the goal stops there, but because it can’t start anywhere else. Once teachers have felt what it’s like to direct an AI build of their own, they’re positioned to hand that same ownership to their students: not as instructors delivering a finished tool, but as guides showing students how to become architects in their own right.

Drawing on real practice at a bilingual K-12 international school in Hong Kong, we introduce “vibe coding”: describing what you want in plain language and letting an AI assistant build a working, self-contained tool — a quiz, a flashcard trainer, a matching game, with no coding experience required. We frame this through four lenses that structure AI literacy at GSIS: learning to create with AI, protect data and integrity while using it, understand what it can and can’t do, and shape it critically rather than accept it uncritically. What ties the four together is agency: literacy here isn’t something students receive, it’s something they practise by building, breaking, and rebuilding, and something teachers first have to practise themselves before they can guide it in others.

In small groups, participants take on that architect role themselves: they vibe-code a simple interactive tool for their own subject and grade level, directing the build through guided prompt templates and two to three feedback loops, refining not just the tool, but their own understanding of what they’re building and why, and how they’d hand that same process to a class. We close with a critical debrief: where AI-generated content needs verification, why “it feels faster” doesn’t always mean it is, and how to keep student data out of the process entirely.