Before AI, the industry sold speed under an older name: rapid e-learning. I wrote about it, and I built under its promises. Faster, more, cheaper per hour. AI is the second act of that same promise, with the volume turned up.
Speed is not the enemy. Pretending speed is design is. A machine that can produce a course in an afternoon has automated production, which was never the scarce part. It has not automated judgment, which always was.
What speed buys, and what it quietly deletes
Faster production buys reach and lowers cost. What it deletes, without anyone noticing, is the friction where design decisions used to get made. The pause where you saw the task did not match the goal, cut the duplicate reading, or realised the assessment tested recall when you wanted transfer. Studies of AI assistance love to report how much more content people made. That is an output number, not a learning one, and the two get confused all the time.
AI can build the course faster. Faster course production is not the same as better learning design.
Judgment frames, not model walkthroughs
This is where the old models earn their keep. Not as steps to follow, but as questions to ask. TPACK asks whether the technology, the content, and the teaching actually fit each other (Mishra & Koehler, 2006). SAMR asks whether the tool changed the task or just digitised it (Hamilton et al., 2016). Point either one at an AI-built module and the seams show up quickly.
Why AI raises the stakes
The honest workload question is not whether AI saves time. It is where the saved time goes. Drafting gets cheaper. Judging what was drafted does not. If the hour you saved goes into producing more instead of designing better, the learner ends up worse off, in front of a fuller course.
The move
Let AI draft. Then put the time it saved into the four failures, not into generating more. Production is the cheap part now. Protect the judgment, because that is the part nobody can hand you.