I have a wonderful new group of eighth graders in my Tech Projects class, and they’ve been hard at work brainstorming ideas, narrowing them down, doing research, and finally picking their best one. At that point, I have them create a project proposal that asks them to step back and think about various aspects of their idea, even though they want to dive in immediately. This isn’t easy, and it doesn’t come naturally.

I ask them to examine internal things like their motivation and what they hope to get out of the experience. I also ask them to detail the resources and methodology they’ll use, as well as the milestones they expect to hit along the way. It’s a lot, but crucial.

It’s a multi-week process, with plenty of coaching and conversations before submitting a final proposal. One student seemed to struggle with this even though I could see they were trying. The answers were short and vague, and I wasn’t seeing the kind of deep thinking and research I was asking for. We talked about it several times, and I thought that sitting down together to help break things down and make them more concrete would help, but it wasn’t quite enough. When I have a bright, engaged student who isn’t delivering the results I expect, often it means that my instructions and setup aren’t working. What’s in my head and the way I choose to communicate isn’t always going to click.

As a result, right after class, I sat down with Claude and did a debrief of the situation, while it was still fresh in my head. I also provided the text of the proposal (with the student’s name removed) and specifications for a 3d model they wanted to print. In less than fifteen minutes, I had a step-by-step guide created just for them, based on their idea, that provided examples, choices for them to make, and a checklist to track progress. Incredible. Claude even identified a problem with the model: it required a filament type that we don’t have. This could’ve ended up miserable if it hadn’t been flagged.

When I sat down to talk with the student about their new guide, it felt as if we were both on the same side of the table, looking at a resource that was created to help them achieve a goal that they’re passionate about. My role is to assist, not criticize or dictate. I asked “so, what do you think?” and the response was positive. So were the results, mostly. It was great to see checkboxes checked and the requested parts of the proposal filled in, but when it came to asking about the next step moving forward, it was “get more stuff done.” Looks like I’ll need to keep being deliberate and specific in my instructions.

This class can be tough for students who work best with clear, concrete steps, because it’s built around self-direction and reflection. Differentiation is a goal all teachers want to achieve, tweaking our curriculum to meet students where they are and help them succeed. Going beyond the basics is difficult in practice, however, and having this new option lets me do it better. If I were on my own I think I’d still make a task list and chunk things down, but I wouldn’t have the bandwidth to create a rich guide infused with the student’s idea, and I probably wouldn’t have flagged the technical problem until much later in the process.

It’s got me thinking of the ways that AI can help me provide a better learning experience for my students. Voice transcriptions are particularly helpful, because I can do quick debriefs in the moment and store them as notes if I can’t sit down for a Claude session right away. Chunking, checklists, and examples are easy to generate, too, if you provide the right context. Having Claude identify potential issues is huge, because it can look at the big picture while I’m focused on something specific.

I’ve also had it improve the wording in my rubrics and instructions to help make them more developmentally appropriate. I have my eleventh and twelfth graders do the same sort of work as my eighth graders, and could use the same rubric for both, but customizing it feels like the better choice. Another thing that’s proven useful is developing customized extension activities for students who want more challenge. Some students rush to the finish line, so I try to design things that help them slow down and fill in their project to make it richer and more complete. Others get to the end easily and want to get more out of their experience, but aren’t sure where to go next.

The key to all of it involves two things: judgment and context. First, I don’t want to turn to AI for help with everything that happens in my classroom. I am careful to evaluate which things it might genuinely be good at doing, and are worth the time. I also don’t want to blindly accept what it generates and call it done. Without some substantive back-and-forth, the results are rarely good.

Next, a big part of the reason why this was both quick and effective is because of the time I’ve taken to build up context about the class, bit by bit, over time. It can access the debriefs I’ve done and the course materials I’ve created, so it knows a lot about my class. It’s all contained in a folder of documents that I access and organize with Obsidian, and I’ve given Claude permission to access all of it. This covers resources, logistics, lessons, and other basic information, and it has my thinking about what works well and what doesn’t. It knows where I want to improve, and what my plans are for the future. That’s how it was able to identify a technical issue way before it would get on my radar.

Part of this setup is how I’m wired. I love organizing and documenting things, but I realize that’s not in everyone’s wheelhouse. If you’d like to get some context that AI can use for your own class, you don’t need to go crazy. Just have the LLM interview you about it, use voice transcription to talk through your answers, and have it organize your answers. The prompt doesn’t need to be anything fancy:

Interview me about my class so that I can create a comprehensive set of context for future conversations. Ask about curriculum, teaching philosophy, school culture, student experiences, resources needed, challenges, and course content, along with anything else that might be relevant. Ask one question at a time, and determine what done looks like so it isn’t an endless interrogation.

Using AI to support student learning doesn’t solve the challenges each student faces, and it doesn’t change the long list of things I have to deal with as a teacher, but it does give me tools and options that I didn’t have before.

This lets me support learners where they are in a much stronger way, and that is absolutely worth the extra effort.