Over the past few months, on tech showcase websites like HackerNews, certain Subreddits, and even GitHub (since it now has its own little “social media” implementation), new projects keep popping up with suspiciously clean syntax, elaborate fun READMEs with emojis and ASCII-diagrams, and wondrous promises about the future.

The reality, unfortunately, is often smoke and mirrors, and it’s indicative of a larger societal shift related to artificial intelligence. WAIT, DON’T CLICK OFF YET - Okay, I know everyone out here is getting tired of hearing about AI, whether in a positive or negative context. However, I genuinely believe it’s worth critically thinking about what’s happening right now, and letting your current experiences guide future decisions. To set the scene, let me tell you a story.

A Story About Decay

I want to talk about one concrete example since it sets up this discussion very nicely. A few weeks ago, I was browsing HackerNews, when I saw a project pop up as a “Show HN” article. For those unfamiliar, think of Reddit “I did a thing” posts, but for software. Like many, many, other projects on HackerNews recently, it was yet another AI agent harness with the already insane subtitle “The open-source Claude Cowork for OpenClaw”. What?

Based on the title alone, anyone who is remotely involved in the AI coding scene might assume this is an AI-generated project - as did I. So I took a closer look at the repository. The README.md consisted of a ton of emojis, a massive number of em-dashes, tables styled to represent features in a flashy way, and ASCII diagrams with completely misaligned rendering. All very strong signals of AI writing. The git commits were also full of messages prefixed “feat:” or “fix:” - something that AI loves to do, given no specific constraints, and that I’ve only seen few developers do consistently, especially on hobby projects.1 This is admittedly a much weaker signal than the others - it’s definitely a valid style for commit messages.

A clean README! Some might say TOO clean...

Now, if you are especially redpilled, you might say something like: “This does not prove AI was used. People write like this.” Okay. Exhausting, but okay. To be sure, I went on the author’s GitHub profile, checked their previous projects - and voilà: None of the observed artifacts, from the git commits to the README writing, were present. So, I went into the corresponding HackerNews thread and posted the following comment:

Genuine question - your README is full of em-dashes, emojis, feature squares and ASCII diagrams - none of which are present in your pre-AI era projects. Why do you expect a potential userbase to care to read something you didn’t even care to write? Seems a bit disrespectful to me.

Unlike every other top-level comment in the thread, my comment did not receive a reply from the author. To another user who more generally asked “why AI slop ranks so high on the front page nowadays”, they responded “What makes you think this is AI slop?”, before, less than an hour later, committing a diff titled feat: enhance clarity and consistency in README content and structure, which removed all emojis, LLMisms like “Here’s why that happens”, and most em-dashes from the README.

Improving the structure of the README by removing all the totally human artifacts!

The author has also since set their GitHub profile to private, making it impossible to track their previous projects, as I had done to investigate the likelihood of AI being used. As of about 2 weeks ago, a few weeks after the project was created, it seems to have been abandoned.

Guilty Until Proven Innocent

This story follows a bit of a trend in my recent activity on HackerNews. I will often call out obviously AI-generated projects or posts, to be met with criticism such as:

One thing I’ve learned recently is a lot guys (like here) have been out here reading each word of a given company’s tech blog, closely parsing each sentence construction.. I really cant imagine being even concious of the prose for something like this. A corporate blog, to me, has some base level of banality to it. It’s like reading a cereal box and getting angry at the lack of nuance.

Provided he reviewed it and checked the readme is telling the users what it needs to tell them - what’s the issue? I’ve found documentation to be one of the better tasks AI can perform and see no reason why not to use it provided a human is in the loop.

Your reaction is worse than the article. There’s no way you could know for sure what their writing process was, but that doesn’t stop you from making overconfident claims.

These comments very nicely map to the three types of typical responses I’ve grown accustomed to getting.

The first one, insinuating that I am in the wrong for caring at all, is some sort of linguistic nihilism that I can’t really address further. If that is your opinion, so be it (although I hope this post might change your mind at least partially).

The second one, stating that it’s fine as long as AI is only used for the “boring” modes of creation, like project documentation or a corporate blog post, is probably the most defensible one. It underlines what AI was meant to be, somewhere, at some point - a quality of life improvement that solves menial tasks. A robot that washes the dishes while we create art. Instead, we got one that creates art while we wash the dishes. The problem, though, is that this point hinges entirely on the conditional clause: It’s fine if the model can either verifiably complete the task in an unbiased and flawless manner (it can’t), or if a human supervises the model and takes ownership of the generated artifacts.

The third type of comment is, in my opinion, by far the most dangerous one. It basically says, “you can’t prove this is AI, so stop making accusations”. It implicitly agrees with the threat model of the second type of comment (unsupervised AI output is bad), but combines it with an overwhelming belief that by our framework of morality alone, we should simply assume innocent until proven guilty, praise the author for their contributions until we see concrete proof of their misconduct.

I think it’s generally admirable to assume people are acting in good faith. However, it also becomes somewhat foolish when the opportunity cost to turn a profit in an unethical manner is far smaller than the cost of doing the morally responsible thing. In other words, if it takes me months to build a properly secure piece of software, in good faith, to satisfy a genuine demand, while taking ownership and responsibility for the created code, but someone with a $100 Claude Code subscription can create twenty new Claude wrappers in a week, the relative chance of encountering said genuine piece of software while browsing randomly becomes astronomically small.

To me, in practice, this means that I actively have to assume that every piece of open-source software that was written by a small team and has no pre-agentic-AI track record is vibecoded and potentially dangerous, whether that’s true or not. By extension, it also means any sort of acclaim I would have once connected with the creation of such a project has been replaced with suspicion and apprehension. This sucks, and I understand why it’s in the interest of both real open source creators and those vibe-coding another OpenClaw Ralph Loop Gas Town Sandbox (Yes, all of these words somehow go together now) for it to be otherwise, but I don’t really see another way forward, personally.

It’s Not Just Software

If you’re not convinced of this philosophy yet, I’d like to give you some more examples. I’m sure most people are well aware of the AI intrusion in creative spaces. Many will remember last year’s disastrous Coca-Cola Christmas ad or the intro to the Winter Olympics. Music is another massive ecosystem. A few weeks ago, an article surfaced stating that AI artists have been displacing real artists in iTunes charts - and if obviously AI-generated music act NOMARKMORE pops up in my Spotify Discover Weekly playlist one more time, I might lose my shit.

Blog posts are also heavily affected. As with code, in writing, a lack of proof often acts as a veil for bad actors, and single posts exist along a spectrum of “LLMness”. From corporate blog posts dropping definitely human bars like:

So we decided to point our autonomous offensive agent at it. No credentials. No insider knowledge. And no human-in-the-loop. Just a domain name and a dream.

…to a blog post about the artificial content flood with 38 em-dashes, several “not x, but y” metaphors and random emboldened text, and sometimes even just an artistic attempt at telling a story that just feels a little pretentious compared to the same author’s natural writing.

The common link here is almost always a lack of disclosure. The way these projects are identified as AI-generated is usually a result of artifacts and imperfections in the output betraying the method of creation despite deafening silence by the creators, rather than “We made this with AI. Please consume if you want.” This has real implications for consumption behaviors, and I’d argue most people producing these artifacts are either acutely aware of this, or at the very least implicitly understand the correlation between the use of AI technologies and negative consumer sentiment.

It’s not a secret by any means that current consumer sentiment leans overwhelmingly against the use of AI in creative technologies. Arguably, that should be reason enough not to actively hide its use, regardless of whether the sentiment is logically sound or futureproof. If 90% of people hated seeing pink elephants in their games, you would probably at least want to disclose that your game features four levels filled with pink elephants, rather than actively disavowing policies requiring such disclosures, as Epic Games CEO Tim Sweeney did - yes, even if the pink elephants made the team output three times as fast. I personally also believe this would actually foster acceptance of AI technologies in the long term. It’d decrease the value extraction from deceiving consumers in the interim though, and that’d look bad on the quarterly sales call.

The Attention Economy

I’m a heavy AI user and I don’t pretend otherwise. Over the past few months, I’ve had Claude write several tools for me: A full-stack photo tagging app for Immich, a Reddit data exporter preserving full conversational context, a set of Ansible playbooks to provision my database servers, and much more. I was a top 3% ChatGPT user in 2025. And sometimes, when I’m making an internal web tool for work, I’ll even let Claude generate the favicon. Yeah. I went there.

However, I have a very distinct line. I don’t take credit for AI-generated work. And I think that’s a really, really critical point. Recently, with discussion around Claude Mythos, people are often riffing on OpenAI stating that GPT-2 was too dangerous to release, ignoring that said discussion was never about AGI, but about a fear of increased fake news, spam, and disinformation. I’d argue that this supposition was not only correct at the time, but has already reached a critical mass in today’s internet, without any AGI having to be involved.

When someone used to produce an artifact on the internet, be that a blog post, a piece of software, an art piece or a musical piece, it functioned as a sort of Proof Of Work, signifying that someone has put significant time into understanding a topic or refining a craft. I think it’s important for everyone to recognize that this fundamental concept no longer applies.

And not only does it no longer apply, people and companies are actively trying to accelerate this breakdown in trust to further their own agendas, with Claude Code skills like Plain English aimed at humanizing Claude’s text output (a skill co-authored by: Claude. Ironic) or even the leaked Undercover Mode built directly in Claude Code, which instructs the model to operate in open-source repositories without disclosing that it is AI.

Encountering an AI-generated blog post right now might be annoying because you realize after three paragraphs that you’ve wasted your time - but at least you realize it.2 Most of the time. It is arguably very likely that you have already been fooled a non-zero amount of times into consuming AI-generated content by strongly instructed models. What I’m trying to prepare for, and what I think others should prepare for, too, is a time when that is no longer possible. When every piece of open source software is a threat vector, every piece of prose written online is likely to be astroturfing, and indie musicians that started producing music after 2025 will have to contend with the constant suspicion that they might be AI-generated, or even just re-recording AI-generated pieces.

There will come a time, not far in the future, when the internet finishes its transformation that was started decades ago by Facebook and co. When not only companies, not only social media platforms, but also independent websites, media aggregators, creatives, and individual actors will fight for your limited time and attention by throwing content tailor-made to appease your personal bubbles at you in the hopes of extracting a few ad dollars. And on a surface level, these platforms will not be discernible from the few real human-produced artifacts that will be left. No RSS, no open-source policy, and no small web aggregator will be able to tell you if the hundreds of “independent voices” who are mad at immigrants, or a social policy, or climate change, or pride month, are real or not, at that point.3 Before writing this post, I signed Vanilla.sh up for a few small web aggregators. When inspecting other blogs applying for said aggregators, it’s already quite clear that some of them are not fully human-produced, but are still given the benefit of the doubt. I am massively in favour of expanding the indie web, but it should be clear to everyone that it’s not a silver bullet against algorithmic noise and artificial content.

And when this time comes, I want you all to remember what’s happening right now. When, in the future, someone online insists that their work is not AI-generated or attacks you for implying otherwise, and the technology has progressed to a point where it’s no longer possible to distinguish artificial from human output, I want you to think back to the time when authors tried to pass off their obviously AI-generated blog posts about AI taking over the internet as human writing. The irony is that as AI writing becomes harder to discern from real writing, it will increasingly feel like a paranoid position to assume AI assistance everywhere, but it will also likely increasingly be the correct position.

Unless things change, in such a time, only getting off the internet and interacting with real humans in real life, listening to one another and considering each other’s grief and hardship will lead to understanding and the preservation of the concept of humanity as a whole. You’ll still be able to use the internet, don’t get me wrong. But putting any faith in the word of someone who’s no longer verifiably using their own voice will be no less foolish than your grandmother on Facebook liking a picture of AI-generated shrimp Jesus with the caption “Amen 🦐”.

It’s good to believe in humanity. It’s foolish to turn a blind eye to what the internet is likely to become, and to continue your current consumption habits in spite of it. I hope this article was able to give you an intuition for the dangers that may lie ahead. What you make of them is ultimately up to you. As always, stay human.