Creating a Human-Centered AI Instructional Design Workflow

What Happens When You Ask AI to Document Your Process

At SweetRush, we’ve been experimenting with AI not as a shortcut for doing the work of instructional design, but as a partner for articulating the work—and one recent project taught us something worth sharing. Like all L&D teams, we have our own approach to instructional design. Our goal is to maximize the benefit of our tool set, and to deliver exceptional results to our learners.

What we landed on is an approach for using our multi-part methodology to build a human-centered AI instructional design workflow: a content analysis process, an audience analysis process, a business requirements process, a learner journey mapping process, and a couple of specialty layers for edge cases. Six components in total, each one doing a distinct job, each one feeding into the next.

Here's the part that matters: We didn’t build that system by asking AI to invent an instructional design framework. We built it by taking work our instructional designer already does, over and over, on real projects—and writing it down precisely enough that AI could run it the same way she would.

Here are six things we learned along the way:

1. Start from something real. Not a hypothetical process you imagine you might need—the thing you actually do, on an actual project, with actual source material. Abstraction is kryptonite to an AI workflow.
2. The AI is a scribe, not an expert. The instruction documents we built read like our designer explaining her own judgment out loud—what she’s listening for, where her inputs come from, what she does with what she hears. The expertise was always hers. The AI’s job was to run a written version of it, consistently.
3. Be precise about output—down to the column names. This sounds small. It isn’t. Every part of the methodology ends with an exact output format: table structure, numbering conventions, and—just as important—what’s deliberately left blank for a human to complete later. Vague output shape is one of the most common, most avoidable failure points in any AI workflow.
4. Build in the judgment boundary. A phrase repeats across every instruction document: ask rather than guess. Flag rather than resolve it yourself. That single design choice does more work than almost anything else in the system—it tells the AI exactly where its judgment ends and the human's begins.
5. Let real use rewrite the instructions. When this methodology was applied to a real client project, it surfaced real friction—undefined terms, naming collisions, an objective that needed more support than the original format assumed. Rather than just fixing the output, our designer went back and redlined the instructions themselves. The workflow isn’t static. It gets better every time it’s run for real.
6. The real test of a workflow. The instructions ended up clear enough—and portable enough—that someone else on the team, or even a different AI instance in a different project, could pick them up and run them consistently. That’s the bar we’d propose for any AI-assisted workflow: it isn’t finished until someone other than its author could use it.

The bigger takeaway, for us, is this: The highest-value use of AI isn’t asking it to do your work for you. It’s asking it to help you write your own process down—precisely enough that it becomes repeatable, teachable, and ready to improve with every real run. This workflow reflects one instance of our ongoing work exploring how AI can support—not replace—the expertise our instructional designers already bring to every project.

AI should multiply your strategic vision—not dilute your content quality. Connect with our team to discover how we can blend expert human judgment with cutting-edge AI to co-create exceptional, high-impact learning experiences.