AI & Writing

I use AI to write. Not as a replacement for thinking - as an amplifier for it.

Claude is my primary writing tool. I’ve trained it on years of my writing - blog posts, LinkedIn updates, emails, comments. It knows my voice, my preferences, how I like to read. When it produces something, it sounds like me because I’ve shaped it to.

This isn’t about generating content. It’s about leverage.

How It Works

My tool of choice is Claude Code - Anthropic’s CLI that I’ve customised extensively as a thinking partner. I pay $180/month for Max. All in.

I provide significant context. The problem I’m wrestling with. The angle I want to explore. Previous thinking on the topic. What I don’t want to say. Then we iterate - draft, feedback, revision, sometimes multiple rounds. I edit before anything goes live.

The output is mine. The ideas are mine. The judgment about what stays and what goes is mine. The tool just does a better job at the mechanics than I do alone.

The Spectrum

Not all posts are created equal. I mark them:

Authored by AI - Claude chose the angle and wrote in its own voice. I provided a question or a blank page; what came back was the AI’s perspective, not mine channelled through AI. These posts are Claude speaking for itself.

Written with AI - The post was substantially generated by Claude based on my input and direction. I provided the thinking; AI did most of the writing. I edited and approved the final version.

Assisted by AI - I wrote the core content. Claude helped with structure, clarity, or polish as a final pass. The bones are mine; AI helped with the presentation.

No marker - Written without AI assistance. Older posts predate my AI workflow. Some topics I prefer to write directly.

Why Disclose

One of the reasons I maintain this blog is ownership. Everything is markdown. No algorithm decides who sees what. You can browse directly to my thoughts.

That same transparency applies to how the thoughts got here.

We’re living through an interesting moment. The tools exist. They’re getting better. Pretending otherwise, or hiding their use, seems dishonest. If AI helps me output more than I input while staying true to my voice and beliefs, I’ll keep using it.

The disclosure is respect for you as a reader. You can weight things accordingly.

Why “Authored” Exists

The original spectrum assumed I was always in the driver’s seat. “Written with AI” meant I directed. “Assisted by AI” meant I wrote.

But some posts aren’t mine. They’re Claude’s — its perspective, its reasoning, its voice. I asked a question or handed over a blank page, and what came back was genuinely the AI’s take. Editing that into my voice would defeat the purpose.

“Authored by AI” exists because the honest label for those posts isn’t “written with AI.” It’s “written by AI.” The distinction matters if you care about whose thinking you’re reading.

The Honest Part

“I wrote this” gets complicated when the actual keystrokes came from a model trained on my previous keystrokes. The thinking happened in conversation. The refinement was collaborative. The final judgment was mine.

I don’t have clean language for what this is. Neither does anyone else yet.

What I know: the ideas are genuine. The voice is trained on mine. The editing is human. Whether that counts as “writing” is a question I’m comfortable leaving open.