LXD Spotlight: James Peters on AI

Join Jocelyn Wright, our Director of Community and Market Development, for a conversation with James Peters, Team Lead, LXD, on how AI is transforming instructional design—from building tools to rethinking the role of the designer.

What drew you to AI-informed instructional design? What do you find exciting or interesting about it?

I got into instructional design through behavioral psychology, working as a teacher in an experimental school based on behavioral science principles. I’ve always been fascinated by teaching machines and by the idea of using technology to reach the masses with teaching, and how technology could improve many problems that happen, mostly in K–12 environments since that’s really where I spent the beginning of my career. With the advent of AI, the ambitions of the early behaviorists and their teaching machines can now be fully realized!

Do you have a favorite LLM? What is it?

My love of different LLMs has changed over the last 3 years. Currently it’s Claude Opus 4.6.

When I first got my first account with ChatGPT in 2023, it was to use DALL-E, the image generator, for an art project I was working on. As these LLMs have changed and evolved, my preferences have changed and evolved. There were things I wanted ChatGPT to do for work a year ago that it tried valiantly and just couldn’t actually do. Then, about 2 months ago Claude released this Opus 4.6 LLM, and suddenly I was like, “Amazing, I’ve underestimated what Claude can do. I need to up the ask!” It’s suddenly become capable of doing things I never ever could see it doing before.

What kinds of things can Claude Opus 4.6 do that weren’t possible before?

One of the most interesting things I’ve witnessed is how these engines are improving all the time. They’re evolving right under our feet.

A great example is that for a long time, I wanted to create some interactive learning objects (ILOs) that could replace the ones we’re building with Articulate Storyline. I had some limited success where I could use Claude and say, “Let’s make a drag-and-drop activity,” and it could do that. I could say, “Let’s make it WCAG 2.1 AA accessible,” and it could do that. And then I’d come back a week later, and it would have forgotten that the ILO needed to be accessible, that the dimensions needed to be a certain way. It couldn’t keep the context.

What’s changed is there’s been a huge jump up in its capabilities in the past few months. I came to Claude one morning recently and said, “I don’t want you to just make an ILO. I want you to make me a tool that makes ILOs.” I went to get a cup of coffee, and I came back, and the basic design was sitting there on the page. And then I started working from there to build extra page types and all these other things.

Now I have this lens that everything can be made as a tool. I don’t even look at a task as “Let’s chat and make something.” I think about what’s the higher version, the abstraction of that task, and Claude is able to produce amazing stuff. I have a tool I just built that will go to any Canvas page and pull the contents into a zip file for download. It’s very helpful if you want to make transformations to those pages separately with another tool.

What have you learned from making tools in Claude?

Claude uses skills—its deeper set of instructions for how it goes about doing complex things. I learned from the ILO authoring tool project that as long as I created a skills doc and attached it to my subsequent prompts, Claude would stay oriented to what it had done before. I recently had a project where we’re producing instructor-led PowerPoints. I took the client’s last PowerPoint sample and their style guide and had Claude generate an idea for a new PowerPoint template. I was able to give the client 3 different template options, all using a color scheme that fit within their branding profile. That’s something that would have taken an entire afternoon just 4 months ago—and now it can be accomplished in 15 minutes.

Another example is developing internal training. We just did a big test run of building 20 courses for a client with 9 different people building them and 4 different leads managing the 9 people. There was a lot of variability in terms of output, so we had a session a few days ago where we looked at a sampling of these courses to identify where we were being inconsistent in our approaches. From that meeting transcript, Claude was able to develop a decision list for us. There were like 48 points of inconsistency we identified in that session. Nobody can take those kinds of notes and put that into a document where these are your options and choose A, B, or C. When we get together for the follow-up session, we’ll use the Claude doc to efficiently make group decisions about all of those items. Then we can pivot back to a group training for the next group of IDs joining the project.

As Claude and other generative AI capabilities increase, how does the role of the ID change? What is the ID still doing when AI can do so much?

Providing the instructional design expertise and anchoring it in a specific context is what we’ll do as IDs in the future. We will not write items and develop materials hands-on; we will edit, analyze, and review content output. We will not be the doers. We’ll be the specifiers. That’s how I see our jobs changing, and that might not be for everybody, but it’s happening. Brave new world.

Just because of how I’m wired, I’m so happy to still be in the workforce and do this while it’s happening. It’s just so endlessly amazing to me. But I know folks who got into this because they love instructional design and development; they love the creation aspect of it. They like the practice of taking the theory and making it into something. I do wonder how those peoples’ jobs are going to change and whether that ultimately will be satisfying for them.

There’s a lot of chatter about AI fatigue. Are you and your colleagues experiencing any of that as a result of the changes to how you’re working?

I was in a call with colleagues a few weeks ago, and we all started talking about knowledge management. We’re all feeling that fatigue already.

Soon, we’ll be awash in content.

For example, Claude can produce a 25-page manual for onboarding team members to a new project, but who has time to actually read and digest it all? That 25-page training document is sort of unusable in the sense that it’s just too much. I worry about that.

AI can be a polarizing topic for some of our clients. How do you navigate AI conversations with clients if you’re unsure how they feel about AI?

I talk to them about the things AI is good for and the things it’s not good for. Usually, if you talk about what it’s not good for, that will resonate with them because they may already have some reluctance. So you already start by saying, “Well it’s not good for the real thinking of the thing,” and they’ll be like, “That’s right!” But then you’ll say, “Well it is really good at creating this or that.” And then you always talk about how everything has to be reviewed; everything must be seen through a human lens. The raw output is not product—it’s only the first step. I also talk about the safeguards we have at SRM to make sure that their content is staying within our walled garden and not being used to train the LLM itself. I’ve found that can reduce resistance.

I think a year ago we were all worried that if we start talking about using AI we’d run into that resistance. What we’ve seen is that our clients have been educating themselves in the past year, like everyone else on the planet. They’re getting more comfortable with the benefits and what it can and can’t do.

AI is constantly evolving and changing. Where should someone who’s interested in learning more about AI in instructional design go?

Go to Claude and ask it to develop a course on AI for you. It is the ultimate teaching machine. It can be used to generate any kind of learning materials you can imagine. Anything you can describe in terms of what you want, Claude can go and make that be.

You can also say, “I need to learn about you. What do you need to know about me?” It’ll begin prompting you—and that flips the dynamic. You don’t even have to think very carefully at the beginning about what it is that you want to do; you can ask it to ask you. Claude will give you menus of choices; it will start branching for you to completely define the thing you are asking it to do.

So I’d recommend that anyone who wants to learn about Claude AI goes and asks Claude AI.

Thank you so much for chatting with me, James!