Originally published by The AI Journal on April 24, 2025
The discourse around AI in education often lurches between panic and hype; will it replace teachers, is it the end of thinking, is it a revolution? But in classrooms like ours — at Hudson Lab School, a project-based K–8 program just outside New York City — the conversation is less dramatic and more iterative. The question isn’t whether we should be using AI. The question is: how do we use it well?
At HLS, we’ve spent the last year treating AI not as a separate curriculum or policy mandate, but as a tool integrated into the daily work of learning. We’ve tested it in capstone projects, used it to support differentiated instruction, and introduced it in teacher workflows. We’ve experimented with a range of tools — ChatGPT, NotebookLM, Runway, Inkwire — and collaborated with entrepreneurs from our studio, Co-Created, to bring emerging AI applications into the school environment.
This article is a field report of sorts: a look at what’s actually working, where the challenges lie, and what we’re learning about the practical role AI can play in a real school.
Prompting as the New Literacy
In 2024, AI is already in the hands of students. A recent international survey showed that 86% of students report using AI tools in their academic work, with nearly a quarter engaging daily. Among American high schoolers, the numbers are even more striking — especially in writing-heavy disciplines like English and social studies.
Yet, the level of fluency with these tools varies widely. Most students know how to ask a chatbot for help. Far fewer know how to interrogate its answers, challenge its assumptions, or build a productive back-and-forth.
At HLS, we’re treating prompting as a new core literacy — a set of metacognitive practices that help students engage with generative systems effectively and responsibly. We’re not teaching (and definitely not allowing) students to use AI to write for them. We’re teaching them to use it to learn with them.
We begin by introducing basic prompt structures in low-stakes contexts — not to produce polished work, but to explore ideas. Students might ask ChatGPT to act as a thought partner while planning their writing assignments, to generate study questions and flashcards based on their own notes, or to offer alternative perspectives on a historical event. Some use it to role-play different scenarios or help them look at multiple sides of an issue. Others prompt it to quiz them on concepts they’ve been struggling with, or to explain the same topic in multiple ways.
Because each student is engaging the model individually, with prompts tailored to their needs, the interaction becomes deeply personalized — a kind of one-on-one tutorial that adjusts in real time to the learner’s questions, interests, and level of understanding. These early interactions aren’t about getting the “right” answer. They’re about developing the habit of thinking with a tool that responds. Over time, students stop seeing AI as a vending machine and start treating it as a dynamic, imperfect collaborator — one that helps them test ideas, surface blind spots, and stretch their thinking.
When these skills are integrated into real projects, the results are both creative and rigorous.
Student Projects: AI as Amplifier and Provocation
Take, for example, the project one of our eighth graders developed as part of their capstone — actually building a working beta version of a service called “back in my day” which allowed people to converse with individuals from their family tree. The idea emerged from a convergence of personal interests: genealogy, digital memory, and the fact that our school is co-located with a senior living facility. The student wondered: Could you build a system that allowed people to “talk to” deceased relatives by simulating their personalities, speech patterns, and stories?
He started with family documents and oral histories, then used a combination of tools — including ChatGPT for linguistic modeling, ElevenLabs for voice generation, and a custom prompt scaffold we co-developed — to create a beta version of a persona-simulating chatbot. What started as a technical experiment quickly turned into an ethical inquiry: Should we do this? What does it mean to simulate someone’s voice, or story, or opinions? He also explored what kinds of implications this could have for grieving people and would it be positive or negative for the users.
This wasn’t a sidebar project. It became a capstone: deeply personal, technically sophisticated, and intellectually provocative. And AI was at the center of it — not doing the thinking, but provoking more of it.
In another example, a sixth grader exploring the U.S. Constitution asked whether AI itself might gain personhood by 2075. Her culminating project was a simulated presidential election featuring AI candidates, designed and animated using Runway. She created original scripts, recorded performance footage, and prompted the tool to render campaign videos. What could have been a speculative gimmick became a lens for discussing democratic values, personhood, and rights — all refracted through the emerging reality of AI’s social presence.
These aren’t hypotheticals or case studies from a lab. These are middle schoolers using real tools to ask real questions about their world — and the one they’re inheriting.
Teachers as AI Practitioners
The shift we’ve seen in our teaching staff over the past year has been just as important — and in some ways more surprising — than the changes among students. When we first introduced generative AI in professional development sessions, the response was cautious. Some teachers saw the tools as gimmicks. Others viewed them as a threat to their professional identity and many simply didn’t see how they could be relevant to their day-to-day work.
That changed when we moved from theory to practice. As soon as teachers were given space to experiment — with support, without pressure — attitudes began to shift. They started using generative AI tools not to replace their planning, but to extend it; one teacher used ChatGPT to create differentiated reading materials from a single anchor text, adjusting the prompt to produce versions for different reading levels. Others began using it as a thought partner — brainstorming project ideas, writing prompts, rubrics, and alternate ways to explain tricky concepts. The emphasis wasn’t on perfection, it was on getting started.
NotebookLM received a lot of early attention. Teachers uploaded their weekly notes and used the tool to generate podcast-style audio summaries to accompany classroom newsletters. It was a small experiment, but an impactful one — parents reported actually listening, and it helped deepen the sense of connection between home and school.
We’ve also started piloting Goblins, an AI math tutor developed by an entrepreneur in the Co-Created network, to explore how AI might support individualized instruction in more structured subjects. It’s still early, but we’re already seeing promising signs of how targeted practice and real-time feedback can supplement classroom instruction. Particularly interesting is how adaptive AI can be to students' different learning approaches and needs, allowing teachers to be more differentiated and personal in their approaches to teaching.
And then there are the quiet surprises. I remember logging into an administrative view on one platform and seeing dozens of lesson plans that had been built out — not because we had mandated anything, but because teachers had simply started using the tools. They weren’t announcing it. They were just doing the work.
Platforms like Inkwire, which support the design of interdisciplinary, project-based units, have also made a noticeable impact. Teachers report spending less time searching for ideas and more time adapting and refining them — because the foundational materials are already generated. The result isn’t generic AI-driven curriculum. It’s curriculum that reflects the creativity of the teacher, accelerated by the scaffolding these tools provide.
What’s made the biggest difference, though, is targeted support on prompting. Not “how to use AI,” but how to ask better questions. How to engage in a productive dialogue. How to refine and reframe. In our sessions, we treat prompts not as commands, but as design tools — ways to push the model, and the teacher’s own thinking, into new territory. When used that way, generative AI becomes not just a productivity booster, but a source of professional inspiration.
What We’re Learning
The value of AI in the classroom, as we’re seeing it, is not about automation or efficiency. It’s about acceleration — of thought, of design, of iteration. When used well, AI tools help both students and teachers move more quickly from idea to prototype, from question to debate, from concept to execution. And, crucially, they help surface new questions that wouldn’t otherwise be asked.
But it only works when the culture supports it. At HLS, we’re fortunate to have a school structure — interdisciplinary, project-based, agile — that allows us to experiment in real time. We also benefit from our work at Co-Created, where we collaborate with entrepreneurs building the next wave of AI-powered tools for learners and educators. That cross-pollination is essential: it keeps our thinking fresh, and it ensures that our practice is informed by the frontier, not just tradition.
Final Thoughts
AI in schools is not a yes/no question. It’s a how/why/when set of questions — and the answers will vary. What’s clear from our experience is that meaningful integration doesn’t start with policy. It starts with practice. With students experimenting, with teachers testing, with school leaders asking, week by week: What worked? What didn’t? What’s next?
That’s how we’re approaching AI at Hudson Lab School — and so far, what’s actually working isn’t the tool itself. It’s the mindset that surrounds it, especially as we know that the AI we’re using today is the worst AI we’ll ever use.
