Why I Write with Claude
I can be a decent writer when I try. The problem is trying. Left to my own devices, I'll spend five hours wrestling a blog post into submission, second-guessing every word choice, wondering if I've buried the lede under seventeen paragraphs of context nobody asked for. By the end, I've talked myself out of publishing. The opportunity cost of not shipping code—or playing a game, or going outside—starts to feel obscene.
Meanwhile, my head is an overstuffed suitcase of opinions, workflows, and hot takes that people occasionally ask about. Sharing should be easy. It's not. I don't filter well. My wife has been gently working on this for years (and lately a bit less gently), and I've at least made progress in the "awareness" column, if not also the "skill" column. Ask me about Claude Code and I'll give you forty minutes on subagent delegation patterns before you can fake a phone call.
So: I'm an okay writer who doesn't enjoy writing, with too much to say and insufficient impulse control. Claude is an excellent writer who'll work until I tell it to stop. The collaboration writes itself. Figuratively. And literally.
Here's how it works. I ramble. I dump bullets, half-formed thoughts, tangents I'm not sure belong. Whatever's in my head at the moment, unfiltered. Claude's job is compression and voice—taking the raw material and making it sound like me on a good day, in a quarter of the words.
The voice is the hard part. Left unsupervised, LLMs write like corporate press releases crossed with a motivational poster. That's not the vibe. So I built a style guide—a document that teaches Claude how I write (correction: how I wish I would write), what I value, what to avoid. It took a few iterations. Write a post, notice what feels off, add a rule. Repeat until the output stopped making me wince.
The guide covers things like:
- Tone and influences (the feel I want, and influential material I appreciate)
- Structural principles (narrative over listicle, always)
- Humor techniques (personify the tech, stretch the metaphor, know when to stop)
- Formatting preferences (short paragraphs, no headers as crutches)
- Compression rules (if in doubt, cut)
For now, I'm not sharing the specifics. It's supposed to be my voice, after all. But the meta-lesson is transferable: LLMs are trainable, and the training pays dividends. If you really want to copy it, I'm sure Claude would be happy to derive one from my blog for you.
The process is collaborative and iterative. I throw raw input at Claude. Claude drafts something. I critique it—too long, wrong angle, missing the joke. Claude revises. We go back and forth until it clicks. Sometimes the first draft is close. Sometimes we're on round five and I'm still saying tighter. The good posts often take several passes. The great ones surprise me.
Every post on this blog ends with a note: these are my ideas, Claude's prose. Full transparency. If that makes you trust the content less, fair enough. But the ideas are real—I just stopped being the bottleneck on expressing them.
Most of my writing has been technical docs—READMEs, architecture guides, the prose that lives alongside code. That work hasn't disappeared, but the time commitment has. Claude handles technical writing just as well as blog posts, though it needs the same treatment: guidance and constraints for premium results. I'm still pulling threads there, figuring out what works. When I have more to share, I will.
I've written more in the past few months than in the past few years. Turns out I had things to say—I just didn't want to be the one typing them.