Designing AI Lessons Teachers Can Actually Use

As artificial intelligence (AI) becomes a bigger part of education, the challenge is no longer whether students should use it, but how teachers can design learning that preserves rigor while building AI literacy. This three-part series explores the research, classroom practice, and practical strategies behind using AI to deepen thinking rather than replace it. 

The most critical lessons we learned were not about AI, but those about instructional practices. Throughout this series, we have argued that the conversation around AI in schools needs to move beyond detection and toward design. In the first article, we explored the research and frameworks that shaped our thinking. In the second article, we examined what happened when those ideas were tested in classrooms across three disciplines.

Those experiences left us with a different set of questions. Which design choices mattered most? What would we repeat? What would we change? And perhaps most importantly, where can teachers begin if they want to explore AI without redesigning an entire curriculum? The lessons that follow are less about technology than they are about teaching.

Our Advice

Perhaps the biggest and most humbling surprise of this work was realizing that none of our lessons succeeded because of the AI itself. The success came when AI forced us to become more intentional in our instructional designs. Technology did not magically make our practices more innovative nor us more effective educators; it exposed the quality of our design and offered us the opportunity to innovate our pedagogy.

If there is one thing this year has taught us, it is that the quality of an AI interaction is set almost entirely before the student ever opens the tool. The AI in our physics project was only as Socratic as the prompt it was built on, and that prompt was only as rigorous as our own understanding of the standards behind it. The tool did not create the pedagogy; it executed the pedagogy we already had to design. This is the part that gets lost in most conversations about AI in schools. The work is not learning the technology; it is knowing your content and your standards well enough to build a task worth automating. The framework comes first, and the tool comes second. This is the practical face of TPACK: the technology was only ever as good as the content and pedagogical knowledge we brought to it, and no amount of comfort with the tool could substitute for knowing our standards well enough to design around them. 

In practice, that means you have five decisions to make as a professional: 

1. Identify the standard you’re accountable for
2. Choose a phenomenon students care about
3. Decide where AI helps and where it steps back
4. Build a checkpoint that students cannot pass without returning to you
5. Hand the work back, watch what happens, and adjust

That is also the reassuring part, because it means the entry point is small. You do not need to rebuild your curriculum or master a new platform. You need one assignment, one phenomenon, one question you genuinely want students to wrestle with, and a decision about where the AI is allowed to help and where it is required to step back. Decide what the student has to produce on their own before the tool engages, build in a checkpoint, and issue the password, literally or figuratively, so the student has to return to you. Then watch what happens and adjust. None of us built our projects right the first time. The most common failure we saw was a task that looked rigorous on paper but let students get through it without committing to their own reasoning first. And the fix was almost always to add an earlier checkpoint, before the AI engages, where the student commits a claim in their own words.

The four of us came at this from very different subjects, and that turned out to matter, because the approach did not stay locked inside a physics classroom. What unites us is neither the subject nor the tool. It is the stance. International school teachers already do the hard version of this work every day. We adapt across languages, across national frameworks, and across students who arrive with very different backgrounds, and we hold a standard steady through all of it. Between us, we have taught in 14 countries on 3 continents, and in every setting, the constant has been the same: knowing the content well enough to keep the bar in place while meeting students where they are. Designing an AI-integrated task draws on exactly that muscle. You are not being asked to become an AI expert. You are being asked to stay what you already are: a reflective practitioner who knows the difference between a student doing the thinking and one watching it happen.

We did not arrive at any of this in isolation, and the wider research is moving in the same direction. In a 2026 systematic review of fifty-four studies, Slimi found that scholarship on generative AI has shifted markedly from a defensive posture built around detection toward what he calls principled redesign, an approach that treats AI as something to be taught with rather than policed against, and that favors assessment built on process, reflection, and the critical evaluation of AI output. That is, in essence, what the frameworks we have leaned on throughout this article already point toward, and what our checkpoints and rubrics were trying to do in practice. The research consensus and the classroom are beginning to meet. Most of that research still lives in higher education, which leaves an obvious next step for those of us in schools: to keep testing, carefully and honestly, what principled AI design looks like with younger students, and to share what we learn. 

The Balance is in the Design

We began by claiming that the balance between AI literacy and academic standards is not a compromise but a design problem, and students' conversations with these AI tools bear that out. The standards held, across AP scope, NGSS practices, mathematics content and skill development, and Sociales standards with rigorous driving questions, not because we restricted the tool and not because we trusted it, but because we built it to demand the thinking we wanted to see. Every question it refused to answer, every checkpoint, every password that sent a student back to a teacher: all of these were design decisions made by people who understood the content first and reached for the tool second. That ordering, in Mishra and Koehler’s terms (2006), is the whole of TPACK: technological knowledge only becomes powerful when it is already in the service of content and pedagogy, never preceding them.

Return for a moment to where one physics student started, to the student who told an AI tutor that space has no gravity. By any traditional measure, he was the student a high-stakes final was most likely to fail: a 1.9, real hardships outside the classroom, and a foundational misconception still intact in May. He was also the student this project served most powerfully. Nobody handed him the answer. He reasoned his way to it, correcting himself, and arrived at language a physicist would recognize. At the checkpoint, he got up and found his teacher. That moment captures the most promising role for AI; not to eliminate the processes that are critical for learning, but to expand the human curiosity and creativity that lead to understanding.

That is what we mean by balance. Not less AI traded against more rigor on a sliding scale, but lessons designed carefully enough that AI literacy and deep disciplinary thinking become the same act. The technology will continue to change and our students will either shape it or be shaped by it. The work of teachers will not. Our responsibility remains what it has always been: to design learning that asks students to think deeply, question honestly, and return, again and again, to the conversations that only people can have. If AI has a place in our classrooms, it is there, not as a replacement for good teaching, but as something deliberately designed to make good teaching matter even more.

This series has shared one school's experience. And brings back what our school means when it asks us to prepare for leadership in service of the world today. We hope it encourages others not to wait for the perfect AI policy or the perfect tool, but to begin with what teachers have always done best: design learning worth thinking through.