Have students log what the AI did for them
After an AI-assisted task, one honest question: which parts could you now do alone? The answer calibrates them better than any rule.
Sep 21, 2026 · 1 min read
Learners systematically confuse fluent performance with durable ability, and assistance makes the illusion worse: things feel learned because they went smoothly (Bjork, Dunlosky & Kornell, 2013). AI is the most fluent assistant students have ever had. A short structured log after AI-assisted work forces the sorting: what did I contribute, what did it contribute, what could I now reproduce cold. (Manning's metacognitive-monitoring-ai-contexts skill targets exactly this blind spot.)
In practice
- Three log lines with the submission: "I did / it did / I could now do alone." Thirty seconds, no essay.
- Spot-check the third line occasionally with a two-minute unassisted redo of one piece (see "Put an unassisted checkpoint inside every big task"). Students calibrate fast once the claim gets tested.
- Watch the trend, not the single log: a student whose "it did" column grows every week is outsourcing more and learning less, visibly.
Quick takeaway: the log turns "it went well" into "here's what I can actually do now."