By 2026, the honeymoon phase with Generative AI is officially over. We’ve all seen, and perhaps been responsible for, the “slop” that happens when you ask an LLM to “Design a course on X.” The result is a statistical average: a bland, generic curriculum that hits the high notes but misses the soul of the subject matter.
If you’ve found yourself burnt out from endlessly correcting AI-generated hallucinations or massaging generic outlines into something usable, it’s time to change the relationship. The secret is to stop focusing on prompt engineering and instead to create a better pipeline.
That fatigue has a cost we do not talk about enough. The AI brain-fried human looks at what constant reviewing of machine output does to a team.
The Inverted Brainstorm: AI as the Interviewer
Starting by asking AI for ideas is the first mistake. When you do this, you are grounding the discussion in the AI’s training data—the “statistical average”—rather than your own expertise.
Instead, use Inverted Prompting. Tell the AI:
“I want to brainstorm a module on [Topic]. Do not give me ideas yet. Instead, ask me a series of questions one by one to extract my thoughts, my unique perspective, and my feelings on how this should be taught.”
This change offers key benefits:
Maintains ownership: The DNA of the project remains yours.
Identifies blind Spots: Once your thoughts are out, you can ask the AI to challenge your narrative.
Forces grounding: It prevents the AI from drifting into “hallucination territory” because it is forced to work within the parameters you’ve just provided.
The "Word Vomit" Strategy
The greatest gift AI offers the Instructional Designer is the death of the blank page. In the past, the transition from brainstorming to outlining required a “clean” first draft. Now, you can embrace the Word Vomit.
Don’t worry about grammar, flow, or professional tone. Type (or dictate) your messy, unorganized thoughts directly into the tool. Use the AI as a high-speed structural editor.
THE GOAL
Take your raw, chaotic insights and ask the AI to format them into a specific structure—a curriculum map, a module breakdown, or a Gagne-inspired sequence.
Because the input was your specific content, the output won’t feel generic. However, be wary of the fact that AI tends to suggest the same activities across different projects. Your job is to “Yes, and…” the AI’s suggestions, pushing it to tailor the interactions specifically to your content rather than accepting generic fit.
The Management Model
The most effective way to view an LLM is as a Supporting Instructional Designer.
As the Lead ID, you are responsible for the final product. The Supporting ID is fast, eager, and occasionally overconfident in their mistakes. You wouldn’t hand a Supporting ID’s first draft directly to a client without a thorough review, and you shouldn’t do it with an AI.
Deciding what the Lead ID keeps and what the Supporting ID gets is a repeatable call, not a vibe. Our Human-AI Delegation Framework for L&D sets out how to make it.
Authority: You must hold the architectural authority. When the AI suggests a sequence, ask yourself: “Does this follow pedagogical best practices, or is it just the easiest way to organize text?”
Verification: If the AI suggests a case study or a technical fact, verify it outside the LLM.
Mentorship: Use the tool to do the legwork (formatting, summarizing, initial structuring), while you provide the mentorship and content, the high-level oversight that only comes with experience.
The Slow-Roll Production Pipeline
Don’t try to automate your entire workflow overnight. That leads to quality debt” that you’ll have to pay back with interest during the development phase. Instead, use a Modular Integration strategy.
The Laboratory Phase
Test new AI techniques on “non-client” projects. Use internal learning challenges or personal side projects to experiment. If a tool or prompt fails here, it’s a learning moment; if it fails on a client project, it’s a disaster.
For example, in our team, we recently built out an automated slide-deck building pipeline as an experiment. Within that same month, we needed twenty slide decks (English and French!) prepared for a client with a tight timeline of two weeks. Using this automated slide deck pipeline, we were able to maintain quality, consistency, and deliver on time, while our designers and developers focused on high-cognitive effort work over manual text entry.
The 'Live' Testing Phase
When moving a tool into a real production pipeline, treat the first few runs as a pilot.
Expect it to break.
Budget time to step in and do the work manually if needed.
Identify the “Scope of Success”: Where does the tool shine, and where does it need constant hand-holding?
Reclaiming the Craft of ID Work
In 2026, transparency is key, but it must be handled with nuance. You should mention that you use AI tools in your pipeline to improve efficiency, but the AI should never be a participant in the room.
Your Subject Matter Experts (SMEs) and stakeholders are paying for your expertise, your empathy for the learner, and your ability to navigate complex organizational needs. By keeping the AI “under the hood” as a formatting and organizational tool, you prevent stakeholders from devaluing your work or pushing for faster or cheaper results that compromise quality.
By offloading the grunt work of formatting and initial structuralizing to your “Supporting ID,” you reclaim your mental bandwidth. You finally have the time to think critically, design more immersive interactions, and solve the complex pedagogical puzzles that actually move the needle for learners.
For the writing craft specifically, a human-centric writing process for instructional design shows what the pipeline looks like end to end.
Our AI Accelerator Certificate for L&D is built around exactly this: depth over speed, with the pedagogy kept in your hands.
