The Julie Kaiser newsletter · no. 2 · August 20, 2026
Buried. Wired. Watermarked.
Hello! So many of you wrote me when you signed up for the newsletter or left a comment on my website. You told me your stories, what you’re building, and what you’re struggling with.
I read every single one and thought, how in the world am I going to email these people back?
That brings me to what went wrong since last time.
What went wrong: I asked you to reply, and then I didn’t answer you
I never expected so many people to email me. And I happen to have two major work deadlines right now and no capacity to write back.
Ironically, the system I’ve built to make my work easier is the one thing I can’t use to make this easier. That’s because I have to follow GDPR, and anything personal and identifying has to be treated with certain precautions. I wasn’t prepared for that.
Your emails are sitting in my inbox. I’m thrilled they’re there. I’ve read all of them. I’ve replied to some of them. If I haven’t replied to you, it’s not because I’m ignoring you. It’s because I’m working through the pile manually, and it’s taking time.
Why this matters: This was a blind spot. I’m careful with the rules on client work, but I’d never thought of my own inbox as the same kind of responsibility. It is. The care I take with a client’s data is the care your email deserves too, even when no one would know the difference.
I am setting up a GDPR-compliant process to speed up my replies. I am not going to send automatic replies, don’t worry. I just want to speed up my drafting, grouping, and idea gathering. I’ll post about this soon, for others in the same boat.
What went right: I finally did a thing I’d been putting off for a while
This month has been one technical thing after another. The garage repeater that never connected, the speakers that kept dropping off my Wi-Fi network, the file mess I told you about last time where my hard drive filled up and kicked my files out to the cloud.
I set up a NAS, a little backup box that’s connected to my network always on, out of the way, and just does the backup thing without me having to think twice. I had no idea how to set up a NAS.
But coding agents have lowered the barrier so now, people who don’t know how to do these things can do these things.
When it comes to technical stuff, there are the people who have no clue and just pay an expert to fix it. There are the techies who just know how, and it’s a fun weekend project. And then there I am, in the middle. I don’t know what I’m doing, but I will read every blog post, every thread, every video, so I can figure it out and do it myself.
But sometimes that’s a really high barrier. It takes a lot of time, so things don’t get done or just sit on the back burner.
Why this matters: Everyone said AI would democratize writing, but what I’ve come to realize is that AI coding agents have democratized doing. They’ve lowered the barrier so I can gather what I need to know fast and have a knowledgeable “assistant” alongside me, guiding me through the actual doing. They really lower the time cost of being a figure-it-out-and-do-it-yourself person.
And that’s really exciting and fun.
What’s new: Claude is going to watermark what it writes
Anthropic says Claude will place an invisible watermark in the text it generates, and people in my streams are in an uproar about it.
It works by skewing which of several equally good words gets picked while the text is being generated by the LLM. The result is a pattern that is invisible to us but readable by people who hold the secret key.
Apparently Google has been doing this in Gemini since 2024. OpenAI says they will do the same. It’s all because of the EU AI Act, which came into force earlier this month, and people are upset about it.
It looks like on one end, people doing manual writing who just put their text into Claude for minor corrections are afraid their own work will be watermarked. That’s fair. On the other end are people pumping out massive amounts of AI slop trying to make a quick buck. They obviously don’t want a watermark either.
I sit somewhere in the middle. I openly draft using gen AI, specifically Claude. I develop drafts with Claude, I react, I think on it, I change it. My work will most likely have a watermark. But I won’t try to strip it or beat a detector to pass off my work as not having been processed by AI.
I’m transparent about how I work and what my text and content actually are. And when I use AI in client work, I make sure we’re on the same page about the deliverable.
So far they see the financial benefit, and other writers in my field do too. As long as we’re providing useful, well-written content that our end reader needs, does it matter that it’s AI assisted?
Why this matters: I think if you’re honest about using AI and the work is genuinely good, a watermark changes nothing for you. It only catches the people pretending they did it all themselves. So don’t pretend. Make the work good, and be upfront about how you made it.
It’s early days. We’ll see. I’m curious to see how this plays out.
So glad you’re here.
Cheers,
Julie