My first course was built at my kitchen table: several textbooks, a PowerPoint from scratch, handouts for every session and an assessment to finish in class. Nobody called it instructional design. It was just the job. Today, a first draft of that whole course could exist before my coffee went cold.
That is the obvious story. Mine is a bit longer. Over nearly two decades I have built learning in classrooms, in large online programs, with universities, and now with AI. If you lead learning at a college, a university or a large organisation, what I took from each stage matters more now, because AI can multiply good learning design as quickly as it multiplies courses.
In the room
I co-founded and owned hospitality businesses for ten years before I moved into training, so I had one test for everything I made: would this survive a real shift? As a trainer, assessor and education manager, I built the learning and then stood in front of the people it was for. If an example fell flat, I knew within minutes and fixed it before the next session. I did not realise it then, but that was the best feedback loop I would ever have.
When you teach it yourself, you know straight away whether it worked.
When the learner left the room
Then the work got bigger. I moved into online programs delivered at scale, and later into partnerships with universities. A course stopped being one person's job and became a process: someone scopes it, someone writes it, someone checks it, all at a distance from the learner. That is how institutions reach thousands of people, and it works. The trade-off is that by the time a real student opens a course, its decisions are months old, and one unclear example in a template confuses everyone who uses it, not just one class.
The further you are from the learner, the longer a problem stays hidden, and the more people it reaches.
When learning left the course
Customer education changed what I think a course is. Some of the best work was not a course at all. It was a short moment placed exactly where someone needed it, like a clear explanation just before a decision. Conrad Gottfredson and Bob Mosher call these the moments of need, and they have argued for years that learning support belongs in them, not only in the classroom. Nobody wanted to take a course about it. They wanted help when it mattered. That is also where I first used AI seriously. It did not replace the thinking, but it removed a lot of the waiting.
The most useful learning arrives at the moment someone needs it, not weeks before.
When it became instant
Now I work with teams adopting AI to create courses, and I see the difference it makes every day. What took weeks takes minutes, and a designer can test two versions of an activity with real people in an afternoon. That is a gift. The time we used to spend building can now go to the parts of the job only people can do: deciding whether an example is believable, accurate for this field and country, and respectful of the learner. I work in English and Spanish, and AI is a brilliant first translator, but a person who knows the culture is still what makes it land. Those are the same judgement calls I made at my kitchen table. Now we get to make them with far better tools, and far more often.
Faster drafts give us more time for the part only people can do: judging what is right.
If you lead learning at scale
Most conversations about AI in education start with which tool to buy. The more useful ones start with what you measure. Measure what learners can do, not what you produced. Courses built and completions are easy to count, but they tell you about output, not learning, and Will Thalheimer's Learning-Transfer Evaluation Model ranks attendance and satisfaction as weak evidence of learning, and real performance as the strongest.
Test with real learners before you scale a template, because the cheapest time to find a problem is before it is copied a hundred times. Name who is accountable for every AI-assisted draft, since accuracy, accessibility, language and fit all need a person who answers for them. And invest in the people who judge. Educators and designers are not the slow part of the process. They are the quality control.
If you design or teach
Whatever the tool, I keep coming back to three things. The first, from Cathy Moore's action mapping, is to start with what the learner needs to be able to do afterwards, not what they need to know. The second is to work out where they will need it, which is sometimes a course and sometimes a short moment at the right time. The third is to test early with real learners, before a problem becomes expensive to fix.
Tools will keep changing. These will not. The courses worth building will still start with someone who needs to do something differently, and be checked by a person who cares whether it is right before a learner ever sees it.
Sources: Cathy Moore, Map It: The Hands-on Guide to Strategic Training Design (2017); Conrad Gottfredson and Bob Mosher, The 5 Moments of Need: A Performance-First Approach (2025); Will Thalheimer, The Learning-Transfer Evaluation Model (LTEM) (2018).