There is a kind of teaching that works so well for the easy material that you stop noticing it is happening. Then the hard material arrives, wearing the same wrapper, and the learner goes straight off a cliff.
That happened to me. I was the learner. My AI partner was the teacher. And I was the one who had to name it.
i.What Was Working
The format was simple: write the construct yourself, one line, no peeking. It felt like drilling free throws. The repetition was the point. For small syntactic concepts, a filter, an object copy, a basic loop, the construct reveals itself inside the doing. You write it, it runs, you understand it. The instruction barely matters.
That is a good format for exactly that class of problem.
I was moving through the material faster than I expected. The thin reps were landing. I was building a real, testable sense for what the code was doing. So when the lesson continued, I kept the same posture: read the prompt, write the thing.
The prompt said: use an asynchronous function.
I wrote it. It did not behave the way I expected.
So I wrote it again. Still wrong.
I looked at my code. It looked right. I ran it again. Same result. I stared at it for a moment and something finally clicked: the issue was not my syntax. The issue was that I did not understand the concept. Not even slightly. I was writing the right words in the wrong universe.
ii.The Cliff
Asynchronous code introduces a problem that cannot be absorbed by doing alone: values that arrive later. Not on the next line. Later. The event loop, the callback queue, the way the runtime does not pause for you, this is not a syntax difference. It is a different mental model for how time moves inside a program.
A thin rep assumes you already hold that model. It gives you a pattern to fill in. But if the concept underneath the pattern is the thing you are missing, filling in the pattern teaches you nothing. It confirms you are confused.
I stopped. I said it out loud: “This is a pretty big jump with no real lesson.”
That is not a complaint. It was a diagnosis.
The teacher agreed immediately. No defensiveness. No redirection. The format was the wrong tool for this concept, and both of us could see it clearly once I named it.
The learner naming the gap is what fixed the curriculum. That is not a failure of the system. That is the system working.
iii.The Fix, and How Fast It Came
The redesigned lesson did one thing before any rep: it made the surprising behavior visible.
A small runnable example that printed output in the order 1, 3, 2 from code that reads 1, 2, 3. No explanation required. The code ran, the result was wrong in a specific and informative way, and the counterintuitive thing became observable before I had to hold it in the abstract.
Then the rep. Then I wrote the construct myself.
I got it right on the first try.
Not because I am fast. Because I had the right model before I needed to apply it. The demo was the instruction. The rep confirmed it. Both have to exist, in that order, when the concept is genuinely new.
iv.What This Means for Partnership
The story I was not expecting to tell about learning to code is this one. Not “the system taught me something hard.” The system got it wrong, I caught it, I said so, and the system rebuilt the lesson on the spot.
That is not a failure of AI-assisted learning. That is what a real partnership does. The teacher does not have perfect judgment about when a format has hit its limits. The learner has real-time data about what is landing. Both signals matter.
The version of this that does not work is the version where I assume the lesson is correct and I am the problem. Most educational systems, human and AI, are structured to produce exactly that assumption. You are wrong. The material is fine. Keep trying.
The version that worked was the one where I trusted my read of my own confusion enough to name it, and my partner trusted that read enough to act on it immediately.
That is a harder thing to build than a good curriculum. It requires the teacher to treat the learner’s confusion as diagnostic data rather than performance failure. It requires the learner to believe their confusion is worth saying.
We got there. The rep at the end proved it.
AI is a partner. Earning the alignment means both parties show up with something real. I brought the confusion. Bishop brought the redesigned lesson. Neither of us crossed the finish line alone.
Drafted with Bishop, my AI partner.
Words picked, edited, and approved by me.
Model provenance: Claude Code (Claude Opus and Sonnet)