I’m writing this from my van. Or rather, I talked this from my van — hands-free, driving, having a conversation with ChatGPT that turned into the raw material for the post you’re reading now.

Let me back up.

Why I’m Blogging Again

For almost ten years I’d been heads-down on client work, busy enough from inbound referrals that having a public presence didn’t matter. Then things got quiet, and I had time to think. I dusted off the website. I started wondering if I had anything worth saying.

Turns out, maybe. I had ChatGPT scan the repos I was working with and asked it where I was ahead of the curve and where I was behind. To my surprise, there were a few things I was doing that were genuinely uncommon — at least in the blogosphere. I’ll get into those in future posts. This one is about how I’ll be writing them.

The Problem with AI Blogging

I wanted to use AI. But there’s so much pushback against AI slop right now, and for good reason. I didn’t want to contribute to that pile of interchangeable, hedge-everything, “in today’s rapidly evolving landscape” content.

The idea I landed on: have the AI interview me instead of having it write for me. That way the voice stays mine. The opinions stay mine. The AI just asks good questions and organizes what I say. In theory, you get the time savings without the slop.

How It Actually Works

The workflow has a few steps.

Step 1: Map out the idea. I work with Claude in Cowork mode (basically Claude with access to my files). Together we flesh out a blog idea file — target audience, key themes, rough talking points, and a set of seed interview questions. Nothing fancy, just a markdown file.

Step 2: Package the interviewer prompt. This is the part I didn’t see anyone else doing quite this way. I have a set of “skills” — markdown files that describe how each phase of the process should work: brainstormer, interviewer, writer, reviewer, researcher. Claude assembles the relevant skills into one self-contained prompt I can copy-paste into another tool.

Step 3: Paste it into ChatGPT and talk. I tried Claude’s voice mode first. It sort of worked — I could speak, but it was slow, and the desktop integration just wasn’t that good. ChatGPT’s voice was better for this. So what I did was have Claude produce one big prompt with everything baked in, copied that into ChatGPT, and when I was driving, hit voice mode, and started talking.

Step 4: Paste the transcript back. After the interview, ChatGPT spits out a full Q&A transcript. I drop that back into my Cowork session, where the writer skill takes over and drafts the post from my actual words.

That’s it. (Well, then edit - it wasn’t quite terse enough for me - I’ll update my writing style guidelines for the next one.)

The Researcher Skill

One thing I added that I didn’t see in other systems: a researcher agent. Mid-interview, if a question comes up that needs actual data — who did this first? what does this tool cost? — the system can pause, go look it up, summarize the answer, and fold it back into the conversation.

For this post, the research question was: who first came up with the idea of AI interviewing a human to generate content? The answer is nobody in particular — it’s an emergent pattern. Elio Struyf formalized it most clearly with his Ghostwriter project (a multi-agent system with dedicated interviewer, writer, and reviewer agents), and that’s where I drew the most inspiration. But the idea of talking to AI and turning the transcript into content has been floating around independently across a bunch of tools and practitioners.

The Hands-Free Thing

I should mention: this interview was done while driving. Phone mounted, headset on, ChatGPT in voice mode, just talking through the questions while I drove. It’s not perfect — but for getting raw material out of your head, it works surprisingly well.

There’s something about talking that unlocks stuff writing doesn’t. You ramble, you say things you wouldn’t have typed. That’s the whole point. The messy transcript is the source material. The writer skill cleans it up later.

What I Don’t Know Yet

This is literally the first post I’ve written this way, so I can’t tell you it produces great content — you’re reading the proof-of-concept right now. You be the judge.

What I can tell you is the process felt right. The interview was fun and engaging and I got through my thoughts quickly. The separation between “me talking” and “AI organizing” seems to work well.

Try It Yourself

The skills I built for this are tool-agnostic markdown files. I’m putting them on GitHub so you can grab them, adapt them, use them with whatever AI tool you prefer. The core set: brainstormer, interviewer, writer, reviewer, researcher, and a writing-style profiler. Each one is a self-contained prompt that any model can follow.

If you try it, let me know how it goes.