Recently I saw someone say that running an AI course is just “harvesting chives” (a Chinese idiom for fleecing people). I found it funny and had nothing to say.
A lot of teachers put real effort into making the learning experience good. Just because you can get something by asking AI doesn’t mean everyone else can get it that easily. Ever heard of the “curse of knowledge”? That’s exactly this blind spot.
Take me. I taught GMAT for 15 years, and only toward the end did it become second nature. The moment a student walks in I can see where the problem is. I design the homework before, during and after class, I customize the whole study plan, and I can diagnose thinking blind spots precisely.
But when I started teaching AI, a lot of it had to start from scratch. Honestly, the product was badly designed at first. I often ran over time, a 1-hour session turning into 3 hours. Pre-class preparation was thin too: things that could have been installed before class got dragged into the class itself, and concepts students could have learned from a recorded lesson, I taught from zero live. Then there was waiting for students to type, waiting for them to install things, waiting for the AI to finish running. All wasted time…
Knowing nothing at the start and wasting time is normal. But I think a teacher who cares goes back and refines their teaching flow and the product spec of the whole course.
So here is my flow now:
First, a short consultation with the student who’s coming for one-on-one, to pin down the scope of what they’ll build and to hand them a blueprint of every future step on the spot. Then I give them the full recorded course. After they finish it, I give them a customized handbook that tells them which prompts to use when they work with AI on their own. There’s a private message channel throughout, so they can ask anytime.
The bigger problems wait for the two-hour one-on-one, where I ask the student to export their conversation history for me to analyze. In class I demonstrate in person: in the same situation where they work with the AI, how I would give instructions, how I would check the results, how I would direct the AI. Along the way I add a lot of techniques, logic and strategy for working with AI.
Students say what they learn this way is very lean. Fumbling around with AI alone, you hit a lot of blind spots and go in circles. But after about two cycles of this method, every student can independently complete the workflow they came with, and they pick up a complete skill set they can use on other projects.
So teaching AI now is really different from the beginning. First, I almost never run over time, because I’ve moved things to before and after class. Second, students know what they’re doing at each step and what comes next, they’ve communicated fully with the teacher, and they know that what they bought is a finished total solution, not “class hours”.
There’s one more thing in the flow that I tell students straight: I record the whole session, give them the transcript, and turn the transcript into class notes for them. Why? Because a transcript is worth more to the teacher than to the student. Think about it: if you have the transcript of every class, you can have AI distill the essence of your lessons, so preparing in the future isn’t so tiring, and you can keep iterating on top of it.
So the transcript helps the teacher far, far more than it helps the student. The student gets one copy, but you have your own transcript every time, which adds up to hundreds.
More importantly, what the student recorded is always your yesterday. The transcripts you keep accumulating help you build your tomorrow faster. What they learn will never catch up with your system.
I think doing it this way is responsible. Would anyone call a course like this chive-harvesting?
So what I think you should actually look at is how much the teacher has put into the course, and whether the course planning and design is complete and carefully made.
So why do I have such a complete structure? Simple: for a while I felt that I was extremely successful teaching GMAT, riding high and doing it with my eyes closed, every consultation dense with information and not a dull moment. But teaching AI, I was back to being the me of 15 years ago, just starting out on GMAT.
So I asked Fable to distill my 15 years of GMAT teaching experience, to find out how my teaching-material specs had been polished all along, and then transfer them in parallel to AI teaching.
All in all, if this is a teacher with experience, passion and the drive to make the product good, why criticize them for no reason?
All I can say is that people who talk from the comfort of the sidelines have it too easy. They deserve a slap.
Want AI working in your own job, not just in demos? See the 3-month 1-on-1 program →
Addendum: the video on running 18 students solo
I never used to think I could coach 18 students on my own and still handle all of it myself: teaching, answering questions, custom study plans, progress adjustments, designing homework, grading homework. Fully personalized service, and more detailed than what I used to give when I only had a handful of students.
With Claude Code, all of that is doable now. It sounds easy, but pulling it off takes some technique: