Why NXGen begins with a plan

AI can build you a course in a few minutes now. It’ll have a title, modules, a quiz, and probably a stock photo of people pointing at a laptop. It’ll look finished.

Whether it’s the right course is a different question. That’s the question I built NXGen Learning Studio to help answer.

An unreasonably short résumé

I’ve spent close to 15 years in learning and enablement. I started in higher education, designing faculty development and online teaching programs. Then I spent years consulting on course design and learning technology for institutions and companies around the world. More recently I’ve been building AI enablement and education programs for both technical and non-technical audiences.

A lot of this goes back to my doctoral work, where I studied how ready instructional designers are to bring AI into their practice. One question from that research never really left me: what would it take for AI to strengthen a designer’s judgment instead of quietly working around it?

Three windows and a thread

My own work ended up being the test case.

Once the AI tools I was using could produce working HTML instead of just text, I started building curriculum assets that way. The one I’m proudest of was an interactive participant workbook. I rebuilt a document-based workbook as a microsite organized around the training it went with, and it stayed current whenever the slides or facilitator materials changed. That had been a real maintenance headache. Participants always had the latest guide, and facilitators stopped chasing old copies.

The results were good. The process was a different story.

I’d be researching in one browser window, talking to the AI in a second, and building in a third. I keep detailed notes and meeting transcripts, so my source material was rich. It was also everywhere, and the only thing connecting it all was me. Honestly, AI didn’t cost me time. It cost me continuity. Every new conversation started with me re-explaining the project to a very fast, very polite collaborator with no memory of yesterday. Keeping the design coherent across all those tools turned into its own little side job, unpaid, with a stress budget.

I’d worked out a process that worked for me. What I wanted was a place for that process to live, so I could stop holding it together by hand. I also wanted it within reach of designers who haven’t spent years building that kind of process for themselves.

The prettiest content isn’t enough

AI is now part of everyday instructional design. In ATD’s research, nearly two-thirds of instructional designers said they’d started using AI tools within the past year, mostly to outline courses, write objectives and draft content. Accuracy and intellectual property were among their biggest concerns. Early studies of AI-generated course designs keep finding the same weakness: links between objectives, activities and assessment that look right on the surface but are thin underneath.

I see the same thing in practice. There’s a lot of AI-generated training out there that’s polished and barely edited, with very little human judgment in the loop. And here’s the part I care about most. As practitioners, we know we can make the prettiest, most aesthetically pleasing content in the world. If it isn’t the right content, supporting the right goals, a learner still walks away unable to do the thing they came to learn. They spent their time, and nothing changed. What good does that do anyone?

That’s why I didn’t build NXGen as a simple course generator, even though it does include AI drafting. How fast you can generate content was never the problem I wanted to solve. The decisions that make a course worth building were.

My first attempt ran backwards

Which brings me to my first try at automating those decisions. I’m told first attempts often run backwards. Mine was very committed to it.

It was a design-document generator that asked about 30 fixed questions over three rounds and took anywhere from 45 minutes to 2 hours. That’s a long time to spend answering questions, but fine, rigor takes time. The real problem was that it only worked once the training materials were already finished. It was, in effect, a very thorough tool for documenting courses after they’d been built. Accurate, detailed, and about one project phase too late.

The irony wasn’t lost on me. Early in my career I planned courses in a backward-design template: a Word document I answered question by question before any development started. Over time, teams kept the thinking but moved faster, toward rapid prototyping. I wanted that discipline back without the slowness. Instead, I’d built the slowness without the discipline.

Thirty-seven questions, then eight

So I tried working backward from something that had already succeeded. I gave an AI agent everything from a curriculum that had gone really well (slide decks, training materials, notes) and asked how the pieces connected to backward design. I also asked what questions a designer would need to answer to reach the same quality on any new topic. That gave me 37 questions in six phases, from the business problem all the way through evaluation and upkeep.

You might be thinking, “Thirty-seven questions? That sounds worse.” It would be. Stay with me.

I also went back through a set of past design documents, and one pattern stood out. Teams almost always filled in the goals and the audience. The sections most often left as placeholder text were activities, assessment and practical requirements. Those are exactly the decisions a strong first draft depends on. In other words, we were reliably answering the easy questions and leaving “[TBD]” on the important ones.

Thirty-seven questions were thorough, but too heavy for a busy designer. So I condensed them into eight open prompts that you answer in plain language, in roughly the order designers make decisions:

1. Why. What problem or opportunity is behind this, and what will success look like for the organization and for learners? Start from the change in performance you need, not the content someone wants covered, and ask whether training is even the right fix.
2. Who. Describe your learners the way you’d brief a new facilitator. Adults learn best when the material connects to their work and builds on what they already know.
3. Do and prove. What will learners be able to do afterward, and what would you accept as proof? Measurable objectives and the evidence for them come before any content. That’s the heart of backward design.
4. Sources. What should be treated as true, and who confirms it? An AI draft is only as reliable as the sources it’s grounded in and the expert who checks them.
5. Limits. What deadlines, approvals and constraints apply? Settle scope and review early instead of discovering them late.
6. Shape. What are you building, how long is it, and where does it fit? Format and length should follow from the objectives, not the other way around.
7. Practice. What will learners practice, where, and where do people usually go wrong? Practice with feedback is where behavior actually changes. Explanation alone rarely gets you there.
8. Finish and follow-through. How should it look, how will it be evaluated, and what happens afterward? Accessibility, evaluation and upkeep get designed in from the start, not tacked on at the end.

Behind each prompt is a short checklist of the details a designer would want covered. The idea is that the AI only follows up where an answer leaves a real gap. Where teams usually leave blanks, it suggests a reasonable default you can accept or change, instead of asking question number thirty-eight. Nothing gets drafted until a person has reviewed the plan.

That being said, the eight prompts are one recipe, not the only way in. Some courses start from a template, some from a manual build, and some from an AI draft. The prompts are there for when you want to do the analysis and design work properly before you jump into development, without spending the hours those old templates used to take.

To be clear, this came out of practice, not a controlled study. It’s what gave the product its shape.

What NXGen does today

NXGen brings planning, editable course authoring, AI assistance and review into one workspace. Fewer windows, one thread.

• The Learning Design workspace keeps the blueprint, goals and objectives, and alignment together. You can review the audience, goal, objectives and proposed pages before building from an approved plan, and lock the parts you want to keep when the AI regenerates the rest. The eight-prompt intake is the next thing we’re integrating here.
• Alignment shows how your objectives connect to pages and assessment questions. It gives you something specific to check, but it can’t make the call for you. A linked question still has to ask learners to actually show the skill.
• AI Assist offers focused suggestions you can review, use or discard. Drafting a whole course with AI is a separate path that gives you an editable starting point.
• Design freedom with standards. You control layout, spacing and presentation, so courses don’t all end up looking the same, while the content stays in line with current accessibility and web standards.

Objectives can now be classified with Bloom’s revised taxonomy, or Webb’s Depth of Knowledge if that’s what your team uses. NXGen checks each objective’s wording for an action you can actually observe, can suggest a level for you to confirm or discard, and flags quiz questions that sit below the level of the objective they’re meant to check. Whether the level is right is still your call. The tool just makes that question harder to skip.

The part where structure pays you back

Here’s my favorite side effect of planning properly: it keeps paying off long after the first version ships.

The more context you bring (well-defined objectives, SME input, rich source material), the better your first draft will be. But say you have a 16-week course built around 10 learning objectives, and somebody asks for an 8-week version. If those objectives, and the evidence for each one, are captured in the plan, that request becomes a sequencing and presentation problem. You don’t have to start over. The objectives stay the same; only the path through them changes. The same goes for adapting a course for a new audience, a new format or a program review.

That’s why I’m building NXGen to support faculty course design and program review, where showing how outcomes, activities and assessment connect is part of the job. It’s also why I want the plan to leave room for different practices. Some teams work from ADDIE, some from backward design, and some evaluate with Kirkpatrick’s model. Making that flexibility explicit is part of where the product is headed.

So… does it work?

The honest answer: it’s early.

I’ve been using NXGen extensively for my own design work, and the biggest change is the one I was hoping for. The work is a lot less chaotic. The planning structure, the AI and the course itself are all in one place, so I’m not carrying the thread by hand anymore. Testing with other practitioners is starting now. A small group of invited reviewers is working in a development version of NXGen, using a Reviewer Mode we built so they can pin feedback directly to anything on the screen. It’s too soon to say what they’ll find, and I’ll share it when there’s something real to share.

None of that is evidence of outcomes, and I’m not claiming any here. My research background makes these questions hard for me to ignore:

• Does structured planning lead to better design decisions?
• Does it cut down on redesigns after a first draft?
• Does it help newer designers work with the confidence of experienced ones?
• Most importantly, can learners do more afterward?

The closest thing to a five-year plan

NXGen Learning Studio is the first piece of a broader learning ecosystem we’re building. My hope for it is simpler than that, though. In five years, I’d love for designers who use tools like NXGen to be more than power users of a product. I want them to feel confident bringing AI-supported, structured ways of working into everything they do, so they spend less time on the grunt work and more on the work that makes the biggest difference for learners.

AI can still build you a course in a few minutes. I’d just rather we spend a few more of them on the plan.

---

Based on my own account of how the platform came together. Product availability is documented separately in the field guide. My influences include Grant Wiggins and Jay McTighe’s Understanding by Design and Cathy Moore’s action mapping, with Knowles, Mager, Bloom’s revised taxonomy, Merrill, Gagné and Kirkpatrick behind the design prompts.

Explore Blueprint · Read the current platform limits