Thinking Through the Machine

AI as a Writing Partner, Not a Ghostwriter

I’m writing this article with Claude right now. Not “using Claude to write it for me” - that distinction matters, and it’s the entire point of what follows.

People keep asking how I use AI to write. The honest answer is: I don’t. I use AI to think. The writing is still mine. If I can’t defend every sentence, every position, every claim - then I didn’t write it. I just generated it. Those aren’t the same thing.

The Lie We Tell Ourselves

“AI helped me write this” has become a socially acceptable way of saying “AI wrote this and I clicked publish.” We all know it. We’ve all read the output. The generic frameworks. The predictable structure. The vocabulary that sounds sophisticated but says nothing. The work of someone who prompted “write me an article about X” and took what came back.

Think about it this way: if you can’t have the conversation without the AI, you don’t have anything worth saying yet. Your job isn’t to extract patterns and slap your name on them. Your job is to bring something the model doesn’t have - your lived experience, your failures, your hard-won perspective on why things work the way they do.

The distinction isn’t whether you used AI. It’s whether YOU are still in the work.

I Learned This the Hard Way

When I first started using Copilot for content, I did exactly what everyone does. I prompted, I took the output, I made minor edits, I published. Efficient. Fast. Wrong.

The feedback was immediate and brutal: “What is this AI-generated garbage?” and “Yeah, I’ve seen this before - it’s Copilot word vomit.” They weren’t wrong. I was producing content that read like a thousand other pieces because it WAS a thousand other pieces, statistically averaged and regurgitated.

But the professional criticism wasn’t what stopped me. My wife did.

In those early days of Copilot adoption, I started using it for personal emails too. Anything that required careful wording - I’d run it through the machine first. After the first couple of drafts I sent her to review, she responded: “Stop sending me your Copilot junk” and “Stop using Copilot and just write the email like you.”

She knows my voice better than anyone. If she couldn’t hear ME in my own words, I’d outsourced something I had no business outsourcing. The professional critics could be dismissed as harsh. My wife couldn’t be. That was the moment I realized I wasn’t using AI to help me think. I was using it to avoid thinking entirely.

The tool had become a crutch for not having to be present in my own work.

Thinking Through vs. Thinking For

The perspective that changed everything: AI is a thinking partner, not a ghostwriter.

The tool doesn’t have opinions. It has pattern recognition trained on everything ever written. When I sit down to write now, I don’t prompt “write me an article about platform architecture.” I have a conversation. I bring a thesis I believe - something I’ve earned through over twenty years of experience. The AI’s job is to pressure-test that thesis. To find the holes. To question “this is weak, I need an example or a paragraph from your experience.” To help me structure the argument so it lands.

What I bring to the conversation: domain expertise from actual implementation, opinions I can defend because I’ve lived the consequences, the “why” behind positions - not theory, but lived experience through failure, and voice - the thing my wife recognized was missing.

What AI brings to the conversation: structure and organization I might not see, pushback on weak arguments, synthesis across domains I haven’t connected, and iteration speed - we can work through ten versions of a paragraph in minutes.

Neither works alone. AI without expertise produces generic slop. Expertise without AI produces the same article I would have written ten years ago, just slower and with less English precision.

The collaboration is the point. Not the delegation.

What This Actually Looks Like

Here’s the workflow:

I come in with a thesis. Something I believe based on what I’ve seen. The AI’s first job is to push back and challenge that thesis. Poke holes. Find the weak spots.

Then I give examples from my experience to support the position. The AI reflects back what it thinks I’m saying. We adjust until the understanding is right.

From there, it lays out an outline or skeleton of the flow. We iterate on actual content, section by section, sometimes paragraph by paragraph. Over and over until we land on something that works.

Then we do another pass together for things we both may have missed. A sentence that sounds like AI. A claim that needs grounding. A section that doesn’t flow.

By the time we’re done, we’ve often gone through thirty or forty versions from where we started.

Right now, in this conversation, Claude has pushed back on my framing twice. First: “How anyone can use AI to write is the most saturated content category online - what makes YOUR take valuable?” That forced me to sharpen the thesis. Second: the suggestion to lead with the professional failure before the personal one, because the sequence matters for how readers experience the lesson.

I rejected some suggestions. I accepted others. I brought stories Claude couldn’t possibly know - my wife’s reaction, my specific Copilot failures. Claude helped me see that the personal story was stronger than the professional one and should land second for emotional impact.

This is what “thinking through the machine” means. You’re still driving. You’re still deciding. You’re still accountable for every word. The AI is the sparring partner that helps you find the weak spots before your audience does.

The conversation itself clarifies thinking. Sometimes it grounds what I already believe. Sometimes it leads me somewhere new.

The Requirement Nobody Mentions

Here’s what the “10 Ways AI Can Help You Write” articles never say: you have to have something to say first.

AI amplifies. It doesn’t create from nothing. Sure, you could prompt it to write about organizational failure without ever having lived through one. But that output? You can spot it a mile away.

The requirement is experience. Opinions you’ve earned. A voice that’s yours because you’ve lived it.

If you’re trying to use AI to shortcut that requirement, it shows. The output has no edge. No specificity. No “I watched this happen” authority. It reads like what it is - pattern recognition without pattern experience.

Go get the reps first. Then the tool becomes useful.

Here’s My Reality

So how do I use AI to write?

I don’t let it write for me. I think out loud with it. I argue with it. I reject most of what it suggests and keep the parts that sharpen my actual point. I bring the experience and the opinions and the voice. It brings structure and pushback and speed.

The article you just read? I could defend every sentence in a conversation without Claude open. That’s the test.

Not whether AI touched the work - it obviously did. The test is whether I’m still in the work. And just like Kiro has transformed this college dropout architect into a developer, Claude has transformed me from a meager writer into someone who can actually publish. I’ve wanted to do these things. I’ve put the time in to learn how to use the tools to make me better. Not to create garbage that nobody will find useful or just plain ignore.

Everyone that knows me will read any of my articles and say to themselves, “Yep, I’ve heard that before. Just not quite as eloquent and less harsh than how he speaks.”

That’s the test.

If the answer is no - if the tool is doing your thinking instead of helping you think - then you’re not writing. You’re generating.

And everyone can tell the difference.

Oh yeah, this article only took me 26 versions to get right.

---

Photo by Nick Morrison on Unsplash