There's a particular kind of demo that looks incredible. Someone types a prompt, an AI generates a course, and thirty seconds later everyone is asking whether instructional designers are about to become obsolete. It's a compelling trick. It is not the same thing as delivering a learning product that survives six months in a real organization.
The first version is the easy part
Imagine the safety procedure changes. Now update the experience in five languages, preserve the approved terminology, make sure the assessment still measures the right decision, retest the mobile experience, and document the new version for everyone who owns the program. That isn't glamorous. Nobody goes viral demonstrating a localization regression test. But that's part of the actual job.
Ask what happens on Tuesday
When evaluating an AI authoring tool, I want to know who can edit its output, what version history it keeps, how it handles accessibility and localization, whether the organization can export and retain the work, and how it fits into the existing review process. I also want to know what happens when the vendor changes its model or its pricing. Traditional authoring tools have weaknesses. Some workflows are needlessly slow. AI absolutely has a role in removing repetitive production work. But replacing a familiar tool with a faster one isn't inherently progress if it leaves the organization with a nastier maintenance problem.
A better test
Run a full life-cycle pilot, not just a build-off. Give two teams the same performance problem and constraints. Have them design, publish, revise, translate, audit and hand off their solutions. Track time, defects, review effort and actual learner performance. That's a comparison worth reading. The question isn't whether AI can produce something that looks like a course. It can. The question is whether we can own, operate and improve what it produces.
Beyond the AI Hype
More Perspectives in this five-part series: