From "Do This Thing" to "Teach Me Everything": How AI Becomes A Learning Partner

The default relationship is backwards

Most people use AI the way they use a vending machine. Put in a request, get out an answer. “Write me a function.” “Fix this bug.” “Deploy this thing.”

It works. But it’s also leaving tons of value on the table.

Here’s why: every time you delegate a task you don’t understand, you become more dependent on the machine, not less. The gap between what you can do and what the AI does for you widens. You’re not building capability — you’re renting it.

I finally broke out of this pattern. I was shipping my personal site (stevereaser.com) with Claude, and we hit a wall: Cloudflare’s API kept rejecting my deploys with a cryptic error about email verification. DNS records were stalling. Auth tokens were expiring silently. At each step, Claude figured out the fix, executed it, and moved on.

I was along for the ride. And it felt great — until I realized I couldn’t have done any of it by myself.

The moment it clicked

We got the site live. HTTPS, custom domain, auto-deploy, contact form, the whole thing. Well, mostly Claude did it.

And then — not even sure exactly why this popped into my head — I told Claude:

”Go back through everything we just did and explain it to me like I’m a smart but very uninformed person.”

What came back was a six-section breakdown of DNS, email authentication, CI/CD pipelines, secrets management, OAuth tokens, and deployment chains — all grounded in the specific problems we’d just solved together. Not abstract textbook definitions. Not “DNS is a hierarchical naming system for computers.” Instead: “You had four TXT records stacked under the same DNS name. Here’s why that broke DKIM verification, and here’s the one-sentence rule that prevents it.”

The difference between theory and experience-anchored explanation is enormous. I’d lived the DKIM failure. I’d watched the auth token refuse to work even after I’d “fixed” the underlying issue. The explanations stuck because they were answers to questions I’d already felt in my gut.

The Three-Phase Pattern

Looking back, here’s the pattern that emerged — and the one I’m going to repeat deliberately from now on:

Phase 1: Build together (let the AI lead)

This is the vending machine phase, and it’s fine. When you’re starting something new and don’t know the territory, let the AI drive. Ship the thing. Get it working. Don’t slow yourself down by demanding to understand every step in real time.

But pay attention. Don’t zone out. Watch what commands it runs. Notice when it hits errors and how it recovers. You’re building a mental map of the territory even if you can’t navigate it yet.

Phase 2: Debrief (make the AI teach)

This is the phase most people skip, and it’s where all the learning happens. Once the thing is shipped and the pressure is off, ask: ”Now explain what we just did.”

The AI has perfect recall of every step, every error, every workaround. It can reconstruct the entire narrative from “blank directory” to “live website” and annotate each decision with why. You’re essentially getting a private tutor who was sitting next to you the whole time, who knows exactly which parts confused you because it watched you struggle with them.

A few prompting patterns that work well here:

”Explain it like I’m smart but uninformed” — this calibrates the explanation level perfectly. You get real concepts without the jargon wall.
”What was the root cause of [that specific error]?” — anchors the learning in lived experience.
”What would I need to know to do this myself next time?” — focuses on actual skills you can learn, not just misc. fixes.

Phase 3: Test and reinforce (let the AI quiz you)

I asked Claude to set up a twice-weekly quiz that emails me questions about the concepts we’d covered. Not generic flashcards — questions grounded in the specific problems we solved. It knows what I struggled with; it knows how to teach.

It tracks my answers, notes what I’ve mastered, and circles back to weak spots. It’s like spaced repetition but for infrastructure knowledge, and the question bank grows with every session we work together.

What I learned about working with AI effectively

Claude also reflected on how I worked with it — which turned out to be surprisingly useful meta-feedback:

Paste error messages verbatim. When something fails, copy the exact error text into the chat. Don’t paraphrase (”it said something about email verification”). The exact text is how the AI diagnoses the problem. Paraphrasing strips out the error codes, the specific API endpoints, the clues that make the difference between a 30-second fix and a 30-minute goose chase.

Answer structured questions with structured answers. When the AI asks you six numbered questions, respond with six numbered answers. Don’t write a paragraph. The structure lets it process your decisions in parallel and move fast.

Say “you choose” when you don’t have a strong preference. Indecision is the single biggest time sink in AI-assisted work. If you don’t care whether the site runs on Vercel or Cloudflare, say so. Let the AI make the call and move on. You can always change it later.

The compounding effect

Here’s what excites me most about this pattern: it compounds.

Session 1, I didn’t really know how DNS records worked. Now I do. Next time I deploy something, I’ll be able to spot misconfigured TXT entries, and ship faster. I won’t need Claude to explain those parts — I’ll be able to skip straight to the new stuff.

Which means Session 2 covers harder ground. And Session 3 covers even harder ground. The human gets smarter, the AI gets a more capable collaborator, and together you move faster than either could alone.

This is the real promise of AI as a tool — not that it replaces your thinking, but that it accelerates your learning to the point where you can think about bigger things.

The bottom line

If you’re just using AI to build things you don’t understand, you’re leaving 80% of the value on the table. The build is the easy part. The learning is the compounding asset.

Next time you ship something with AI help, add 15 minutes at the end:

Ask it to explain what you just did.
Ask it to quiz you on it later.
Watch yourself get smarter.

The AI doesn’t get tired of teaching. Take advantage of that.

Steve Reaser helps small businesses put AI to work. He writes about what’s actually working, and what isn’t. Watch me learn and build in public at SteveReaser.com