AI makes everything faster. Including garbage.

One of the most seductive things about AI is speed. But we're also starting to see the flip side during this experiment.

AI makes almost everything we do faster. Research that used to take days, we do in an hour. A first draft lands within minutes. For a new product image, we no longer need to book a photographer, studio, and product setup. And analysis of dozens of options sometimes appears on our screen in seconds.

Fantastic.

Except it creates a new problem: when you can produce almost unlimited content, when is something actually good enough? We're running into that more and more at co-Founded by AI.

Thirty seconds is tempting

For this experiment, Jurriën and I are trying to build a product company where AI does almost all the execution. No employees, maximum four hours per founder per week, and €200.000 in revenue as the first-year target. AI researches, writes, codes, analyzes, and designs. We try mainly to set direction and make decisions. That sounds efficient. And usually it is.

But there's a strange temptation lurking. When Claude or ChatGPT produces something in thirty seconds that looks pretty solid at first glance, it's incredibly tempting to think: done.

Marginal cost approaching zero

One more prompt. One more variant. Ten more images. Twenty more headlines. The marginal cost is practically nothing. So why stop? This is what's now being called AI-slop: massive amounts of cheap, generic, and often empty AI-generated content. Or put less politely: garbage.

The Financieele Dagblad recently wrote about how Spotify, YouTube, LinkedIn, book retailers, and other platforms are having to increasingly armor themselves against this flood. No wonder. One person can now produce more content in an afternoon with AI than a creative team produced in a week just a few years ago.

The natural quality brake disappears

We're producing it too. The uncomfortable part is we're using the exact same technology. More than that: without AI, this whole experiment couldn't exist. Take this blog. AI helps us organize our thoughts, track what we've done over the past week, explore angles, and draft a first version. That saves enormous amounts of time.

But then comes the work we apparently can't automate yet. We read every sentence. Is it right? Would we say it this way? Is a conclusion actually based on what we've experienced, or does it just sound good? We strip out typical AI phrasings, fact-check figures and sources, and add examples only we know.

And then we rewrite. And again. Before you know it, we're two hours in on a piece AI could have churned out in minutes. On paper, pretty inefficient for two founders trying to prove you can run a business almost entirely on AI. But that's exactly where the core lies for us.

Let AI do the making

AI takes over production. Not responsibility. Seven times the same cube. We see that maybe most clearly with Minagi. AI can create beautiful product images. In seconds, our focus timer sits on a minimalist desk in Tokyo, next to a coffee cup in Copenhagen, or in a beautifully styled living room.

Until you lay the images side by side. Then Minagi suddenly has a different button. The ridges are different. The wood type changes. The proportions don't quite line up. Sometimes AI spontaneously invents an extra feature we never designed.

We've written before about seven times the same cube. It turned out to be surprisingly hard for AI. The image is beautiful. The product is wrong. And that difference becomes only more important once we move from fictional renderings to an actual product.

Professional-looking nonsense

We hit the same wall even harder with technical design. We asked AI for technical drawings, CAD-like designs, STEP files, and material specs for Minagi. The first results were impressive.

Until our product designer looked at them. Dimensions didn't match. Parts didn't fit together. Production details were made up. AI could explain very convincingly how something technically fit, when in reality it was unmakeable.

We actually find that a more interesting form of AI-slop than a bad LinkedIn post. Because bad AI content you often still recognize. It becomes dangerous when the output looks professional. A spreadsheet with neat figures. A technical drawing full of exact dimensions. A market analysis with charts. A legal-sounding document. A research report with 138 pages.

The more convincing AI becomes, the more critical it is to know when not to believe it.

Making gets cheap

This is changing how we see our own role during this experiment. We started co-Founded by AI partly from the question: how much work can AI take from entrepreneurs? By now we think that question might be the wrong one.

AI can actually take on enormous amounts of work. The real question is: what work is left for us after that? Probably less making and more judging.

• Write less ourselves, but get better at spotting a good story.
• Design less ourselves, but sharpen our eye for taste and consistency.
• Analyze less ourselves, but better understand which numbers matter.
• Less hands-on work, but actually more decisions to make.

That sounds simple. It's not. Because once AI gives you twenty options instead of three, you still have to know which ones to throw out. Maybe that becomes one of the most important entrepreneurial skills in the AI era.

Not getting even better at prompting. But learning to stop. Knowing when something's right. When it's differentiated enough. When to bring in an expert. And when that rapidly generated output just belongs in the trash.

AI makes making nearly free. Good judgment becomes more valuable than ever.