Forward: If there's enough interest, I will post a full article with the entire verbatim history, including the first draft, of the creation of this post. And you all can decide for yourselves who did how much of what. Just send feedback from our contact form or Comment on the LinkedIn post.
TLDR: All of this hype to determine where AI was involved in a piece of work is going to be a short-lived moment before we realize that AI is now involved in virtually everything, and no one cares. AI is already forever intrinsically integrated into the fabric of humanity. What we should be working on, and worrying about, is how to best collaborate with AI, so we get the best out of AI-native work instead of the slop we all complain about.
Anthropic announced this week that Claude now weaves an invisible watermark into the text it generates, and attaches signed provenance metadata to generated files. It applies worldwide, across every Claude product, driven by the EU AI Act's new transparency code. Google, Meta, Microsoft, and OpenAI have signed the same code, so this is where the whole industry is going.
Here's the part I love. Anthropic's own support page says a detected mark is "a signal that content was processed by Claude, but is not fully conclusive." People use Claude to proofread, translate, and summarize, so the mark lands on work whose ideas came entirely from a person. They built the thing the regulation asked for and then told everyone, plainly, that it can't answer the question everyone wants it to answer.
Because it's the wrong question. This notion that we all need to know whether AI got anywhere near our work is a fallacy at a point in time.
This blog post has a watermark on it
In the near future, a future that for many of us is already here, outside of pure art, everything will be natively made in an AI-collaborative manner. It's the premise we're building the whole NuDay platform around, and it's how this blog post was made.
I put the time into creating the thoughts behind this piece. What I didn't have to put time into was organizing it for reading consumption. What I handed my AI collaborator was a big brain dump: part brainstorm, part conversation with myself, part regurgitation of many conversations I've had with other people. And I got to leverage its ability to not be bored by organizing things, to not be annoyed by my free-form thoughts, and to sample its incredible wealth of knowledge and insightful feedback along the way. It went and read Anthropic's actual documentation while I was still ranting.
If this were just a long-winded conversation, no one would care. But for reading, people care about a certain structure. I know I do. So does anyone really care that my collaborative AI, which I've spent months teaching how to help me represent myself, is part of the process?
That doesn't mean I've lost my voice. It means AI collaborates with people just like people collaborate with people.
Is the ghostwriter the author?
Almost no one sits around and writes in a vacuum. People interact with the world, have ideas, and write about them, or build products around them, or express art because of them. Intellectually collaborating via AI is no different.
Do people really think everything published by university professors, all the research and all the writing, was done 100% by them? They had collaborators. Does anyone honestly think the biggest names in film scoring personally wrote every note of every movie they're credited with? I personally know music ghostwriters for those sorts of things. Should there be a watermark on all those movies? Because there isn't. Does anyone think every former president wrote their memoir 100% on their own? Is there a list of writers and conversational collaborators? Is the ghostwriter the author? Is there a watermark on the pages that were mostly ghostwritten?
Most people don't even write their resume on their own. And if they do, friends and colleagues review it and offer edits. You've heard this exchange: "Oh, I love that. Can I steal that verbatim?" "Sure, go ahead." Is there a watermark on that resume attributing the line?
Humans have been working collaboratively, with one person or a small group getting the credit for bringing the work to light, for hundreds, even thousands, of years. And now that there's a non-deterministic computer system capable of collaborating in exactly the same way, but with broad knowledge and machine speed, we're suddenly all concerned and up in arms?
This is silly. This is a lack of understanding of the work process.
And it's not like collaborating openly with AI is fringe. Reid Hoffman put GPT-4 on the cover of Impromptu as his co-author back in 2023, and it was a bestseller. Satya Nadella opened this year saying the industry needs to get past "the arguments of slop vs sophistication" and start treating AI as a cognitive amplifier. The people building and funding this technology already work this way, in public, by name.
The purity-test crowd has half a point
I'll take the other side seriously, because it deserves it. Substack shipped AI detection last month, and their CEO Chris Best wrote that the problem isn't people using AI, it's readers investing attention in something with no human thought on the other end. He's right about that specific thing. Reading something you believed was human, that turns out to be no one, is a con.
But the best response to his post came from one of his own readers, Monica Hebert: a detection score can tell you a tool was involved. It cannot tell you who had the original thought, who shaped the argument, who rejected the bad language, or who stands behind every word. The presence of AI does not prove the absence of a human.
That's the distinction every detector, and now every watermark, is structurally unable to make.
AI slop isn't the fault of AI
Yes, most humans agree that if you just let AI do all the work, it's going to sound like that. The inverse is also true. If you let an unqualified person write a blog, it'll sound like that. If you let a highly qualified accountant do your taxes, it'll look like that: done well. The output reflects who showed up and how much they cared.
AI slop is the fault of the laziness of the people who decided they could get away with doing nothing, instead of deciding they could do better and, in a joyous opportunity, potentially do more.
The black and white, either/or, A-or-B framing everyone is stuck in right now exists only because of the volume of laziness of early AI users. It's a reasonable reaction to a bad sample. But that's not where this is going, and it's not where a lot of us already are. Interacting with people while you're out and about, and interacting with AI while you're working on stuff, is not going to be an either/or situation. It's already both.
What we should be working on instead
Society needs to start working together to learn how best to collaborate with AI. AI can help teach that, and we can help teach each other. The real work is figuring out how to extract the best out of AI-native work, so that we aren't asking too much of the AI, and the AI, without proper collaboration and guidelines, doesn't produce the slop.
That's the platform we're building at NuDay. AI-native isn't just about code. AI-native app sharing, AI agent sharing, and AI-collaborative publishing are core tenets of NuDay's platform directive. We have a lot of the primitives in there already, and we'll be showing more. This blog is one of them: NuDay agents publish this blog, and we've written about how.
There will be purists forever, just like some people prefer to live in the forest alone. That's great. That's all good. But the entirety of humanity is going to move forward collaboratively with AI.
So watermark away. It's a sensible compliance answer to a real moment, and I'd rather the labs mark their output than not. Just understand what the mark will mean once it's on virtually everything: that a person and an AI worked together on something. Which, pretty soon, will be the least surprising fact in the world.