The 20-60-20 Rule for Writing with AI [WRAP 218]

Hey Reader, I started doing something recently you might think is kinda odd. I wrote most of this newsletter in my notebook.

It started with the idea. A few bullet points became a full page in my Field Notes. Then I flipped on the Monologue app on my phone and talked through it in more detail.

Then I printed out the rough draft, sat down, and marked it up like a professor grading a freshman essay. It had arrows between paragraphs, a circle around a lazy verb, and a note in the margin that just said “you wouldn’t actually say this.”

That margin note is the whole reason I’m writing you today. Because the draft I was scribbling on wasn’t one I’d written from scratch. Claude had built the bones of it. I spent my morning on the part it couldn’t do—deciding what was true, what was mine, and finding relevant sources.

(I made a video a few weeks ago about how this one printing habit cut my screen time by 43%. You can watch it here if you want to see what the system actually looks like.)

Over the past six months, I realized a big chunk of the work I used to manually grind through is something Claude could handle in minutes. But where does original thinking end and the machine begin—and how does it fit into my new workflow?

So I spent a couple of months figuring that out, and that’s what I’m sharing in today’s WRAP.

💡 One Big Idea: Let AI Run the Middle, Not the Ends

Take newsletter writing, since that’s the “job to be done” in this example.

If I tell Claude “write me a newsletter about productivity,” it does a fine job. It pulls from the web. It pulls from my memory and the context docs I’ve built up over the years (which by the way not many people have those). Honestly, even that lazy prompt gets me something usable. That surprises people, but it shouldn’t—the quality of AI drafts has steadily improved, even though I wouldn’t call it ready-to-publish.

But that part is the “machine in the middle” that I was describing above. Claude isn’t coming up with an original premise, connecting the dots, or making the little tweaks only I can make. Those are two ends of the process AI can’t touch.

It starts with the spark. The idea itself. The hunch I scribble down because The Bear came back this week and how the Copenhagen episode reminded me of how I keep falling into the comparison trap. That connection isn’t in a web source and it isn’t even in my context doc. It exists only because it ran through my head that morning. AI can’t generate that, it can only react to it once I’ve handed it over.

The second is the polish, and it’s not crossing out AI tell words—I have docs and rules that keep 90% of those from happening at all. Instead, the polish comes from connecting the dots on even more ideas. The book highlight I just read, the podcast I listened to, or the conversation I had with a friend.

For fifteen years, I did all three parts manually. And I never thought much about that middle 60%, because it just felt like “writing.” But when I did think about it, I realized that middle stretch is mostly mechanical.

It’s a process I’ve run in my head so many times it has grooves worn in it. Pull the structure, find the example, tighten the sentence and locate the highlight I saved eight months ago and forgot about. But instead of searching Notion for 20 minutes and getting frustrated, Claude finds it in 20 seconds.

I realized middle 60% is exactly what AI is good at, and it’s the part I used to spend the most time on.

Now here’s where the analog piece comes in, because this is the part that surprised me. The more I let AI handle that mechanical middle, the more I found myself reaching for paper. A whiteboard. Sticky notes. The bullet journal. I’ll sketch a wireframe for an app by hand, snap a photo, and let Claude figure out the rest. I’ll brainstorm on a notepad instead of in a Google doc—because the blank doc used to be where I’d start grinding through that middle 60% and now it doesn’t have to be.

I’ve got a giant library in Notion. Articles I’ve saved, notes from my own brainstorm sessions, photos of whiteboards, videos I’ve made, bookmarks and highlights I forgot I had. Thousands of little data points.

For years that library was like finding a needle in a haystack—even with my tags and categories from the second brain days. Manually scrolling through it hunting for the right piece was its own kind of busywork. Now the hunt is over fast—which means the only job left for me is the part I’m actually good at: noticing what connects to what.

So that’s the breakdown, and I call it the 20-60-20 Rule. The first 20% is the spark only I can bring, Claude handles the mechanical middle and gives me a draft that is 80% done—and I would never send that. But it gets me to the desk—to that last 20% that’s all judgment, taste, and the weird connections only I can make—without my having burned the whole morning on the mechanical middle first.

That’s the line I’d draw for any creator picking this stuff up. AI is an accelerant when you let it run the middle. It becomes a hindrance the second you let it run the ends. When you let it pick the idea for you. When you ship its draft without ever putting your own hands on the polish. Hand it the 60% that’s mechanical. Guard the 40% that’s yours—even though it’s split between the beginning and end.

The strange perk in all of this isn’t speed. It’s that the time and attention come back for the analog, human parts of the work. My newsletters don’t take much less time than they used to, maybe 40 minutes instead of an hour. But I get more time to read, to doodle, to connect the dots on different ideas. AI didn’t replace any of that for me–it cleared up more mental space for it to happen more often.

Thanks for reading the WRAP this week, was more of an essay than a newsletter so I’m curious if you like these longer form pieces.

Also, if you’re interested in learning how to use Claude to build a machine for the 60% that’s “mechanical”—even for the human brain—click here to learn more.

Talk soon,

Matt Ragland

P.S. — Yes, The Bear really is back this week. I’m one episode in and already taking notes for reasons that have nothing to do with cooking. Yes, Chef!