Years ago I inherited an online senior English course built for learners in remote northern communities. The first version was a wall of text. To finish one task, a learner had to read a page here, leave for a video hosted somewhere else, open a document in a third place, and find their way back — with no rest built in and no line that ever said, plainly, do this next. The discomfort it produced was not productive struggle. It was raw load.
The redesign changed almost no content and fixed almost everything. Lean text. One place to work instead of four. A small, deliberate break built into the rhythm. And, above all, explicit prompting: at every step, a plain instruction for what to do next. Afterward, the “what do I do?” messages mostly stopped, and learners told me the course finally felt like it respected their time.
Courses fail at “what do I do next”
A course rarely fails because the content is wrong. It fails at the seams between tasks, where the learner is left to guess. Every guess costs attention that should have gone to the material. This is not a motivation problem dressed up as a design problem. It is cognitive load, and it is measurable: working memory is narrow, and when the design spends it on navigation there is less left for learning (Sweller, 1988; Sweller et al., 2011).
A twenty-minute reading with no task is content. A short reading followed by a decision, an attempt, and a worked example is design.
The twenty-minute learner is the real learner
Design for the learner you actually have: tired, time-poor, learning after a shift or between other duties. That learner does not need a richer map. They need a shorter, better-marked path: density over volume, sequence over coverage, one clear action at a time. Backward design names the discipline — decide what the learner should be able to do, then build only the path that gets them there (Wiggins & McTighe, 2005).
Why AI raises the stakes
AI makes the unclear path more common, because it makes content cheap to over-produce and pad. A generated module will happily hand you sixty screens where six were needed. The design judgment — what to cut, what to sequence, where to place the one instruction that unblocks everything — is exactly the part AI cannot supply, because it does not know your learner’s evening.
The move
Walk your own course as the learner. At every transition, ask: does the next action name itself? If you have to think, they will have to guess. Mark the way. In my own redesign, the single change that mattered most was the cheapest one: telling the learner, every time, what to do next.