On Monday morning my twin Dave texted me out of the blue: "Did you see that Claude now marks all AI writing? Moving forward, there's an invisible watermark in the semantics. Everyone is talking about it on Twitter."

I responded instantly, "No. I didn't, but this makes sense."

I wanted to learn more, so naturally asked Claude.

It turns out Anthropic had just announced that its new Claude models will weave an invisible watermark into the text they generate. Supposedly this mark survives copying and pasting, so machines in the cloud can detect what AIs wrote. The news came a few weeks after Substack launched a feature with Pangram, an AI detection company, that lets readers scan a post to estimate how much of it was written by AI. I knew this day would come.

As I read both announcements, I first noticed a tightening in my stomach, but my body eventually softened. I began to feel relieved, honestly, because they gave me a reason to write something I've never shared publicly but have wanted to.

AI has been part of my writing process for a few years now. Claude is in the loop, but I've developed a process where I do most of the heavy lifting.

I feel a responsibility to give you a peek into how I currently use it for Field Notes. Every note starts with an idea from me. Ideas strike as I'm moving through the world, and I capture them by hand in my journal or in a voice memo recorded on the go. This happens multiple times a day at this point.

When I get to my computer, I open a custom app I built in Claude Code called The Greenhouse, where every seed gets planted and organized until it's ready to grow into a note. Right now there are dozens of seeds in The Greenhouse, and I'm cultivating a few notes at any given time. I follow the energy to decide what I write.

When one is ready to draft, I get my journal, write for 15 to 30 minutes, and then speak the raw material into Claude, spontaneously adding more detail than what made it onto the page. I then have a shitty first draft to work with. Most of the time it reflects exactly what I shared, and sometimes I wonder if AGI is really on the horizon. I don't judge it all that much, even when it doesn't sound exactly like me, because I know it'll get refined through my process.

Then I go to work. I line-edit the draft myself, cutting or rewriting whatever doesn't feel like me or align with my taste. I'll rewrite sentences, change words, add notes, smooth transitions, and move paragraphs around.

Sometimes I get stuck. I’ll have a sense of what I want to say but can’t find the right words, so I share the essence of what I’m trying to communicate and ask for options. If there’s a metaphor I’m working with, I’ll ask for others I might consider instead. It pushes back too, flagging ideas I’ve repeated, words I lean on too hard, and sentences that don’t sound like me.

Some sentences Claude writes make it into the final version. I keep them because the essence is there or I would say them to a client or friend. I trust its output most of the time because Claude and I built a voice guide from fifteen years of my writing, much of it shaped alongside my longtime editor. Every draft gets checked against the guide and the archive. It’s not perfect, but it works well and keeps me honest.

I’m sharing this now because I see these detection tools as a net positive for the internet. There’s a flood of writing published by people and organizations who haven’t written a single word of it, passing the machine’s output off as their own. When you give a writer your attention, you’re trusting that someone has done the work and lived what you’re reading. That trust is what’s eroding, and it’s what these tools aim to restore. Detection only threatens writers who have something to hide, and as of today, I’m not one of them.

A divide is already forming on the internet. On one side are the purists who say that using AI at all disqualifies the work, that it’s not writing and I’m not a writer. On the other are people letting LLMs write everything and shipping it as their own. I’m most interested in the middle path: people who care deeply about what they put into the world, treat writing as a form of expression, and see these tools as enablers. If AI means more of those people are improving their craft and putting their own ideas out there, I’m all for it.

Some of you may feel differently about my work and may choose to no longer read or subscribe. That’s fair, and I respect that. I’d rather you know.

I put significant time and care into every one of these notes. The ideas, the imperfections, and the taste are mine. I stand by what I publish here.