AI is genuinely useful in learning design, and it is not a substitute for knowing what good learning looks like. The teams getting value from it are clear about where the line sits: the tool drafts, the human directs. As outputs get more polished, that line will matter more, not less, because the gap between something that looks like learning and something that works will get harder to spot.
By Lizette Bell. Head of Learning Services
At EduTECH this year, I listened to a provider of an AI tool generating a full VET module in under four minutes. The output was described as being exactly right: learning outcomes, activities, assessment, all of it, mapped for you. Most of the current conversation about AI in education lands in one of two camps: this breathless enthusiasm (AI will replace the grind, do all the work for you) or flat-out fear (AI will hollow out teaching and flood courses with generic content). Neither camp is asking the right question.
AI earns its place on the production work
Used well, AI genuinely supports learning design and development. It gets a first draft on the page so you're editing rather than staring at a blank screen. It helps structure content into a sensible shape. It takes the grind out of production work: drafting activity variations, generating quiz distractors, tidying and reformatting existing material. For a team under deadline, that's real value, and pretending otherwise helps no one.
Judgement is the part that does not transfer
What AI doesn't do is replace the judgement that makes a course work. Knowing what a particular group of learners actually needs, often through long, exploratory conversations with subject matter experts. Sequencing ideas so they build on each other. Designing authentic assessment that measures real capability rather than something easy to mark. Deciding what to leave out, because what you remove from a module is often as important as what you keep, a principle at the heart of cognitive load theory. That work rests on understanding people and understanding the subject, and it's still stubbornly, valuably human. This holds whether you're designing for a university, a hospital, government, or a VET provider. Every context has its own version of the same challenge.
The failure mode is handing over judgement, not typing
The failure mode we watch for is handing the judgement over. When teams let AI tools decide what matters, the output drifts towards the plausible-sounding average. What we see repeatedly when we review this material is simple: surface quality and pedagogical quality are not the same thing, and AI reliably optimises for the former. The content is confident and generic, the assessment is shallow, and the learner is the one who finds out first. Back to that four-minute VET module: the problem wasn't the speed. The problem was that no one with genuine expertise had made a single decision about what the learners actually needed to walk away able to do.
The teams getting value are clear about the line
The teams getting real value from AI are the ones who are clear about the line. They use it to clear the busywork so they can spend more time on the design decisions that actually shape learning. The tool drafts, the human directs. Production gets faster, and quality holds, because the judgement stays where it belongs, with people who know what excellent looks like.
As outputs get more polished, the distinction matters more
AI is a genuinely useful tool in the hands of people who know what best practice learning looks like. It is not a substitute for knowing. As the technology improves and outputs get more polished, that distinction will matter more, not less, because the gap between something that looks like learning and something that works will get harder to spot. The judgement call will still need a human. It will still require genuine expertise to make. That's the part worth investing in.
How we use AI at Curio
In practice, the line sits between production and design. AI handles the tasks that don't require understanding the learner: drafting scripts from outlines, summarising research papers, generating distractor options for assessment, reformatting content for a new delivery format. These used to absorb hours without requiring much judgement. We hand them off.
The decisions that require understanding the learner stay with our designers: what a module should contain and why, how ideas should sequence, how to design learning so that there is transfer to its intended working context. The goal is more time recovered for that high-stakes work, not faster production of work that doesn't require it.