I once ran a course where the feedback was genuinely good and arrived genuinely too late. Thoughtful comments a learner got after the unit had closed, when nothing they did next could be shaped by them. The comments were true, well written, and useless.
A rough note that lands while the learner can still act on it does more than a polished paragraph that lands when they cannot. Not because faster is better, which the research does not support, the studies comparing immediate with delayed feedback are genuinely mixed (Shute, 2008). The wall is actionability. The loop of attempt, feedback, revision is not a feature you bolt onto a course. In a real sense it is the course.
Design the loop, not the comment
Feedback fails when we treat it as something the marker improvises at the end instead of something the design builds in from the start. Boud and Molloy put it plainly: feedback is a loop to be designed, not an episode to be delivered, and it only works if the learner gets a later chance to use it (Boud & Molloy, 2013). Hattie and Timperley add the three questions any loop has to answer. Where am I going, how am I going, and where to next (Hattie & Timperley, 2007).
Feedback that arrives too late is decoration. Design the loop, not the comment.
Feedback as a designed object
The fix that worked for me was not to ask the marker for more. On the course I keep coming back to, the teacher who would mark the work was often hundreds of kilometres away, and the loop through them ran days long at best. But someone was usually in the room: a mentor, a family member, a learning supporter, someone in the building. Not a teacher, not trained in assessment, and never to be handed a rubric written for professionals and wished luck. So the loop was built for them instead. One page, print-safe, checkboxes pointed at the process rather than the person, and three conversation items the form could not be completed without. The expertise sat in the design, which left the human free to spend themselves on the relationship.
Why AI raises the stakes
AI can generate feedback instantly, which is both a real gift and a real trap. Fast, formative, revisable feedback on a draft is exactly the loop we want. Feedback that quietly writes the revision for the learner closes the loop by removing the learner from it. The design decides which one you get.
The move
Find one point in a course where feedback arrives too late to use, and move it earlier, with a built-in chance to revise. One well-placed loop teaches more than three tidy summaries.