The first online course I was ever asked to rescue had everything in it. Dozens of readings, a shelf of videos, three discussion forums, a glossary. It had been built by someone careful, and it was failing anyway. A learner opened it, scrolled, and asked the only question that matters: what do you actually want me to do?
That course was not unusual. It is the most common failure in adult learning, and it is almost never a content problem. The material is accurate, current, often well written. What is missing is the design: the learner cannot tell what matters, what to do next, or how they will know they are making progress.
Coverage is not a design
For years I measured courses by coverage. Every topic addressed, every resource linked. Completeness felt like rigour. It was the opposite. A fuller map of the territory is not a trail through it, and a busy adult does not need the territory. They need the route.
The research has pointed this way for decades. Adults arrive with experience, immediate goals, and limited time; they engage when learning connects to a real task and disengage when it does not (Knowles et al., 2020). And across the design models, the same principles keep resurfacing: learning improves when it is anchored in real problems, when prior experience is activated, when a skill is demonstrated and then applied (Merrill, 2002). None of that is about the amount of content. All of it is about sequence and use.
Content was never the product. When information was scarce, assembling it looked like the job. It never was.
The four failures
Strip away the jargon and every broken learning experience fails in one of four places. Path: the learner cannot tell what to do next. Practice: there is nowhere to try the skill before it counts. Feedback: no one tells them how the attempt went, or it arrives too late to steer anything. Transfer: nothing carries into the work on Monday. Content is never the fifth failure. A pile of accurate material fails all four at once, because a pile has no direction.
Why AI raises the stakes
AI can now generate a full curriculum in an afternoon, and most of what it generates will teach very little. Not because the content is wrong. Content was never the bottleneck. The world has been full of accurate material for twenty years, and if exposure produced competence we would all be experts in everything. Faster production mostly means bigger piles. The trail is still hand-built.
This is the liberation hiding inside the threat. If content was the job, the job is gone. If design was always the job, then the part a machine can flood the zone with was never the valuable part, and the designer who understands that becomes more valuable, not less.
One test before you build or buy
Ignore the topic list. Ask instead: where, exactly, does the learner practise, and who tells them how it went? If the answer is clear, the course has a spine, whatever its production values. If the answer is silence, no amount of added content will fix it, because the problem was never missing material.