Part 3: Make AI Your Student and Escape the Intermediate Trap — Graduating from "I Thought I Understood"

Series: “Careers & Skills in the AI Era” (5 parts)

1. How Not to Let AI Take Your Career
2. Beginners, Build Experience with AI
3. Make AI Your Student and Escape the Intermediate Trap (this article)
4. Grow Leaders and Managers with an AI Team
5. Those Who Grow AI Win

Introduction

Last time, we talked about beginners building experience with AI.

Once you build experience, you reach the next stage: the intermediate. You basically get it; you can move your own hands. But here is where many people hit a wall. This time, we tackle getting past that intermediate wall by “teaching AI.”

The idea of treating AI as “co-intelligence” for learning is popularized by Ethan Mollick (“One Useful Thing”).

The “I Thought I Understood” Wall That Stops Intermediates

Learners split roughly into three stages.

• Beginner: Asks AI questions and moves forward on the answers.
• Intermediate: Basically understands and can move their own hands.
• Advanced: Uses AI as a tool and maximizes results.

The problem is the intermediate. Intermediates are the group most prone to “I thought I understood.”

They can build something that runs, and give a plausible explanation. But when asked why it works, under what conditions it breaks, or why they chose this over the alternatives, it suddenly gets vague. This state of “seeming to understand but unable to put it into words” is the true nature of the intermediate wall. This “I thought I understood” is also called the illusion of competence in research (arXiv).

The Way Past the Wall Is to “Teach AI”

What works astonishingly well for getting past this wall is “teaching AI.”

That teaching deepens understanding has long been known (Learning by Teaching). What makes AI excellent is that it will partner with you anytime, any number of times, endlessly. And you can let it push back without hesitation: “Why?” “What’s the premise there?”

When you try to make AI understand something, you’re naturally forced to explain the following.

• Premises: What situation are you assuming?
• Constraints: What must be observed; what must not be done?
• Criteria: Why can you say that choice is “correct”?
• Edge cases: How do you handle the cases that don’t go smoothly?

The moment you try to put these four into words, the parts you didn’t actually understand rise clearly to the surface.

AI Is a “Mirror of Your Own Understanding”

This is the biggest point.

AI won’t move the way you intend when your explanation is vague. If your instruction is fuzzy, a fuzzy result comes back. When an off-target output returns, it’s usually not a lack of ability in AI—it’s a sign your explanation was incomplete.

In other words,

AI becomes a mirror that reflects the resolution of your own understanding exactly as it is.

When a person is your audience, they read between the lines and fill in the gaps, so you never notice the holes in your explanation. Because AI doesn’t fill them in, it honestly confronts you with the holes in your understanding. As a learning device, this is an unbeatable feedback loop.

This Is Not “Prompt Engineering”

I don’t want this misunderstood: this is not about clever prompting.

What you’re doing is converting your own tacit knowledge into a form anyone can reproduce (explicit knowledge).

• In software, it’s like distilling the design philosophy in your head into a spec, a DSL, or review criteria.
• In work, it’s like turning vague experience into “a way of working that anyone can reproduce.”

“Teaching AI” just happens to have AI as the audience; in essence it is training that structures your thinking.

In Practice: A Five-Step Routine for Teaching AI

Since abstractions are hard to act on, here’s the procedure I keep in mind. When you want to relearn something, teaching AI through these five steps deepens your understanding a notch.

1. Write the context: Make the premises, background, and cast explicit.
2. Write the spec: Define what, how far, and in what form you want to achieve.
3. Write the goal: Put into words what “done” ultimately looks like.
4. Write the success criteria: Decide concrete standards for judging “it’s done.”
5. Review: Read AI’s output and pinpoint where in your explanation the gap originated.

The point is step five. When the output is off, don’t end at “AI is useless”—always pull it back to “where was my explanation lacking?” That round trip raises the resolution of your understanding directly.

And you don’t have to write all five steps by hand. You can draft and organize the context, spec, and goal documents together with an AI agent, and even have AI review them—letting it check a first pass before you make the final call. Let AI do the manual work; you focus on the judgment: what to convey, and what is correct. The learning effect still comes through in full.

Summary

• Intermediates are prone to the “I thought I understood” wall.
• The strongest way past the wall is to “teach AI.”
• Teaching AI forces you to verbalize premises, constraints, criteria, and edge cases, exposing the holes in your understanding.
• AI is a “mirror of your own understanding” and an excellent feedback device.
• This isn’t prompt technique; it’s “structuring your thinking”—turning tacit knowledge into explicit knowledge.

Next time, we move to how this “ability to teach AI” connects directly to the ability to develop people—management and leadership. The stage is the “AI team.”

👉 Next: Grow Leaders and Managers with an AI Team