I made this █████ (Nope: no “with AI”)

Whenever I show one of my AI concoctions — one I’m particularly proud of, or one I want to publish — to my friends, colleagues or acquaintances, something strange happens. I hear myself saying, more or less:

“I made this with AI.”

Somehow apologetically. Pre-emptively, taking the wind out of the sails of any possible criticism, I justify my achievement. Even, or especially, when it was a lot of work and I put a particularly large amount of energy into it. As if it only turned out this good because of the AI, and my share in it were vanishingly small or unimportant.

And if I don’t say the sentence myself, it almost always comes from someone else, from a different direction. Along the lines of: in principle you could do it just as well yourself, you just didn’t have the leisure, or didn’t waste the time, to invest it in that bit of AI chit-chat. A bit of prompting back and forth isn’t an achievement worth honoring.

AI-dissed, there I stand. And part of me probably agrees. The feeling of not really having made it myself. The AI elephant is standing in the middle of the room. So let’s paint it pink and cut it into slices.

Privately and on the side, I develop exclusively with AI. From front to back, the whole Anthropic suite up and down, garnished with Google’s Gemini or NotebookLM, plus a voice generator and a video-specialist tool here and there, or some frontier or low-cost special-purpose instrument.

With that I’ve built a container website consisting of 8 free tools, a prompt framework, an online Erklärbär and an irregularly published newsletter in 6 languages. Lately I’ve also been haunting Insta (de/en) and YouTube (de/en) — for the fun of tinkering with videos and to promote my two bestsellers, Pretty Prepped Prompts (PPP) and Ego-Check. All of it with total devotion and modest success. I’ll leave the trailblazing failed attempts, the duds and all the half-finished projects unlisted, but that’s another 20 projects — depending on how you count.

All part of my AI journey.

With my pseudo-productivity roll call I don’t want to make an impression — OK, not only — but to make it plainly clear that I could never have pulled all this off on the side, alone, in 6 months, without massive use of AI. Not even remotely would that have been possible in the pre-AI era. You couldn’t even have thought of imagining it, that’s how absurd it would have been. And now it has happened, and it runs, scales, is fun and somehow feels right. That still doesn’t make it a walk in the park, and it isn’t effortless either. It costs time and energy. But when it’s fun, those don’t get debited to the burden account — they get credited to the flow allowance.

So when I’ve built a program after three weekends of intense vibe coding — yes, I still code AI-old-school, only vibe and not 100 % agentic-autonomous, and I still have to pitch in myself and readjust the broken loop, and the automated tests only deliver 97 % coverage, and I definitely want that context shift gone, and now that I see the result I’ve got another idea and I don’t like the implementation, so I redesign and redo large parts of it, newer and better — so when I’ve spent my free time producing the program or bringing it to a new version, then I made it. Not Claude Code or Gemini. Me. Period. No asterisk. No little star. No footnote. No apologetic, justifying look. Me — all by myself.

And that’s not self-overestimating arrogance, it’s simply owed to the fact that AI has become standard and AI is a tool I use. Not necessarily virtuosically or uniquely, but not just a prompt copied without love either. And if it’s nothing special, then there’s no reason to point it out.

When I pick up bread rolls at the bakery, the baker doesn’t say:

“Here are your rolls, I made them with the oven and the dough-kneading machine. I learned to bake and I know how to read and follow a recipe. The apprentice helped me. And luckily I don’t have to make the flour myself.”

Maybe he could even grind the grain himself and knead the dough by hand. Replacing the oven, though, would be a matter for the fire code. And above all, he’d bake fewer rolls.

And my very own non-AI share — the idea, the bureaucracy around it, the testing, signing off and refining, the love of detail, building up the kit so that tomorrow I can still bake a bread roll … er … videos and programs — that shouldn’t be underestimated. And even if AI helps me with my own share too: it doesn’t do it of its own accord. Or not yet. Or not that well yet. Or not yet in a way I’d know about or could use. But even if the AI does 95 % of the legwork in the venture, the decisive 5 % stays with me — the part only I can do and only I did.

And so there’s no added value in with AI … no useful extra information. It merely puts a frame around your own work. Defines the toolset and the stage.

I framed my AI affairs long ago — and reframed them again and again. Built and refined a framework, in the truest sense of the word. Set up a small artisanal AI manufactory with industrial traits. The effort behind it alone doesn’t count, of course. The result is what’s relevant. And that it can always be done better and faster isn’t up for debate either. The point that matters goes right after “I made this.” Full stop. No AI label necessary.

I haven’t had this self-image for very long myself, and I rarely see it in others either. And when I look at all the various “I made this with AI” apologies and “You made that with AI, though” put-downs and other AI-soaked, almost Newspeak-like phrases and constructions, I see a lot of superficial profundity in them, and some things that are downright worrying. Care for a few examples?

The effort downplayer — “That took me 3½ minutes with AI. Two more minutes and I can automate it.”
Guilty. I do it too. It’s just too tempting, but also pretty mean. What you’ve spent months learning and building up and have already applied dozens of times naturally comes easier to you. But then making it look as if everyone should be able to adopt it just as easily and in no time at all is unfair. There’s some showing off in it, but also a bit of the curse of knowledge — the cognitive bias that no longer lets you easily imagine the world the way it was when you were a beginner yourself. When you yourself wrestled with the AI for 3 to 9 hours until you finally managed a half-baked and most certainly unreproducible result. Pinky swear: I’ll cut back. Or, like the angler and his sooooooo big fish, it becomes part of accepted AI culture. AI fish stories, then — or AI yarns, for the sailors among us.

False modesty — “I wouldn’t have been able to without AI”
Self-devaluation is out of place. It sounds humble, but it wrongly makes you small and hands over the authorship of your own work. As if you were saying: “Obviously I wouldn’t be able to get this done so well. Sorry for using legal aids and properly deploying the toolset available to me to achieve an optimal result.” That’s probably induced from outside, implicitly or explicitly, through social sanctions. Because there’s no real reason for this attitude. Or there shouldn’t be.

Devaluation from outside — “Well, you did that with AI”
Not accepting other people’s work because it was created with the help of AI. That’s pretty perverse. After all, it means you’re actually satisfied with the result. Probably even highly satisfied. But you don’t appreciate it, because the other person made it easy for themselves — or because you can’t do it yourself. But hasn’t that always been the goal? To get tasks done better and faster through routine and experience? And no sooner has someone taken exactly that to the extreme than it’s suddenly supposed to be worth nothing.

The liability hand-off — “The AI checked it”
Checking is good. And checking with the help of AI is — when done right — a relief, faster, better and therefore effective. And you can’t always convey the tone or the attitude behind the statement well in writing. But sometimes the real core behind it is the dishonest handing-off of responsibility for errors that slipped through the AI’s checking grid, or could have. When you use it to weasel yourself out of responsibility, quite unashamedly, that’s worrying and should be exposed, not silently accepted. It mustn’t wither into a mere formality. If you’re responsible for the content, then checking it — however you do it — makes sense in your own interest. AI or no AI.

The later excuse — “The AI can fix / tidy it up later”
True. It can, or better: you can, with it. And if it actually gets done, that’s a lasting quality gain. But it also likes to get forgotten, or used as an excuse to no longer exercise the necessary care. Why should you still structure data when the AI copes pretty well with the unstructured garbage too? The AI will muddle through somehow, so you don’t have to bother creating and maintaining structure. But that’s a boomerang. And for everyone who sees no value in clean data as such: with structure the AI is also better and faster, hallucinates less and burns fewer tokens.

Incapacitation — “The AI can do it, so I don’t have to be able to”
Subtly dangerous and, in the long run, the most dangerous. Incapacitation by announcement, and painless only at the beginning. The problem isn’t handing a task to the number-crunching drudge that can do it better than you. That’s fine, and with the calculator, the satnav and the phone’s contact list we did that too, without any really lasting damage. Even if, all in all, we’ve gotten worse at mental arithmetic, map reading and finding our way, and at memorizing long numbers. The big difference with AI is that we don’t just lose the one skill itself, but also the judgment to evaluate the result and the system. That’s because AI doesn’t just replace a single isolated island skill like adding, routing or remembering, but complex systems where the connections need to be understood too. For the 95 % of the work the AI takes over to be the right 95 %, we have to be able to do, understand and penetrate at least the AI’s share well enough that we can genuinely evaluate it.

---

Without AI still works, of course. Pointedly emphasized, it comes in two variants: out of defiance or out of misplaced pride. First, the variant without any pride at all.

No-AI resignation — “I made this without AI”
Made entirely without AI. Deliberately, and it’s so obvious that you really wouldn’t need to point it out. The translated meaning: “I didn’t even bother to spit out a prompt so that it would halfway meet the standard of a presentable result.”

The leap from

• not daring to use AI, via
• not trusting the AI you use to
• using AI out of laziness and on to
• not using AI out of laziness

happened surprisingly fast.

In a certain way that’s honest, but also very resigned. You don’t even heat up the canned ravioli anymore before eating — you just put the can down, cold and half-open. Even though there’s a complete kitchen with a microwave and a stove, and a full fridge and a pantry. Still, you ignore all of it and focus on the fact that it fills you up anyway, it doesn’t have to taste good, and you’d have to wash a plate afterwards too. Understandable, relatable, but sad. On a meta level you’re admitting: you obviously haven’t managed to make the hurdle of using AI low enough for yourself that using it pays off and isn’t a burden — or you think this ever-so-small extra effort isn’t worth it, because it’s only canned ravioli anyway and always will be. And presumably is meant to stay that way.

As I said: without AI also works with your head held high and with pride.

The handcraft manifesto — “Solved by hand”
My favorite AI-human phenomenon is the second variant, when it comes along as a memorial and a call to praise the intended imperfection, because it’s still genuine — computer-aided, mind you — handcraft. That’s exactly the sham value of Grandma’s scratchy sweater, knitted with love and effort but unrivaled in ugliness. Or of the homemade tiramisu that, through the consistent replacement of every ingredient, mutates into a low-fat-quark-fruit-agave-syrup-chia layer cake. Nicely uncreamy and unsweet — and it keeps growing in your mouth as you chew. Yes, out of politeness, and when there’s real love in it or the person genuinely means well or can’t do or know any better, you give appreciative praise. Not really for the result, but for the effort and the good intention behind it. But with an Excel sheet that looks handmade and is formally correct, yet has a bit of a higgledy-piggledy character, and where even its creator would know how to turn it into a little gem, I no longer have the strength these days to approve of it.

My conclusion from all the with AI and deliberately without AI: Don’t join in the habit of naming AI as co-author and co-creator in the first place. And if at all, then just like with googling, as a verb of its own. That way it can be used without fuss, without provoking amazement, applause or rejection: to AI. Is that a regular or an irregular verb, then? I AI’d a presentation. Or I AI’d one up? Lowercase and minus the apostrophe: I aid a presentation — which, come to think of it, is closer to the truth. Let me just quickly AI what would be grammatically correct there. And as soon as everyone says it, it disappears again. Who still googles these days? People search again, or do research, or for a brief while they just AI it.

The wordplay is only a gag on the side. What I mean seriously is the share we should keep as humans. Whether that’s 5 % or 17.93 % doesn’t matter much either. AI takes work off our hands, a lot of it even. But the human-AI combo also gets more done overall. The AI has long been doing more than legwork — but not the decisive part that gives the whole thing direction and meaning in the first place. We have to make sure that it is, and stays, the right 95 %. What helps me with that, at least, is grappling with the true meaning behind the facade of AI phrases.

My first step: with full fervor and at the top of my lungs, I’ll keep quiet to my fellow humans about the AI fact of my work. And rhetorical fake-AI questions that aren’t asked out of genuine interest in the how but are hidden put-downs — I’ll expose them as such.

“I made this.”

“You? More like Claude, right?” (a slight grin, as if I’d been caught cheating)

“Me.”

And what about this official AI label, actually? Since August 2026, the AI Act’s statutory labeling obligation has applied to certain AI content.

The opposite of good is well meant.

At least everywhere that visible, unambiguous notices have to be stuck onto content — with the grand promise of protecting us from deepfakes and creating trust in digitalization. That a chatbot in a conversation has to come out with it immediately and proactively that it consists of lots of zeros and ones is 100 percent legitimate and genuinely creates transparency. And that AI content gets invisible metadata under the hood, so search engines can weight it better and the next AI training doesn’t stew in its own juices, copying itself over and over, makes absolute technical sense.

But a visible little stamp on images, videos or texts that’s supposed to work as a weapon against lies and fraud? Toothless and off target. And it certainly doesn’t create trust. For me, it inevitably conjures up the dozens of organic seals on the supermarket cucumber. Which one was the good one again?

I don’t want to be bamboozled by deepfakes either. I don’t want to have to constantly doubt, while watching the news, whether the president really said that or whether it was just deceptively realistically foisted on him. Although with some presidents you fervently hope right up to the end that it’s a bad AI joke. But a mandatory AI seal doesn’t solve the problem of potential deception with AI. And I don’t have a patent solution rattling around in my drawer either.

But we’re making a massive category error here. We’re confusing the tool with the truth content. Just because something has once been pulled deep through an LLM or an image generator doesn’t automatically make it more false or worse. If we have to plaster AI content across the board with warning labels like cigarette packs, because potential fakes might be inside, then we get exactly the opposite of trust: we stigmatize the tool itself. And whoever then honestly labels their true, carefully researched, factually correct but AI-generated content suddenly reaps skepticism for it. Rightly so, perhaps, but amplified by the state and for the wrong motives.

And then there’s the small matter of enforcement. Who dutifully sticks the label on? The good guys. The ones who obey the law anyway. The fraudster and the trolls studiously leave it off. And when — or if — everyone has gotten used to the idea that anything artificial would have to be labeled, the bad guys have an easier time with their unlabeled fakes. A false sense of security created and the honest punished. Bravo.

1896: the first filmed train races across the screen toward the audience, and people are said to have run screaming from the hall. Kept in the dark by the light of the projector. Ironically, even this panic anecdote is, historically speaking, probably a wild exaggeration by the press of the day. An early analog double fake.

The tricks got better and we got better at spotting the tricks, and the whole thing started over again. When I look today at the old special effects I was wildly excited about back then, I do wonder how I could ever have taken a man in a rubber suit for Godzilla. Or why I didn’t see back then that the laser beams were simply scratched onto the film. But here, too, the frame plays an important role. In the feature film frame we know we’re being played and we want that, too — after all, it’s flesh-and-blood actors lying to us there. On the news or out in the wild, it gets trickier.

Only — why do we act as if you needed AI to bend reality? Photomontages have always existed, good ones and bad ones. Photoshopped rolled off our tongues fluently, and still it never earned a statutory warning banner back then. Today an image is simply retouched with AI assistance, or rendered entirely from scratch. But isn’t a deliberately misleading camera angle, a cropped-out context or a false location at least as manipulative?

There’s a famous example from the old Guardian TV ad: a man seemingly runs aggressively toward a passer-by. It looks like an assault. Only from the other camera angle do you see that at the last moment he’s yanking him aside from falling debris. Not one pixel of it was AI. And no law requires a label for that kind of thing: “Caution, framing may deliberately present reality in a manipulative way.”

The desire for truth is absolutely understandable. But a technical provenance label is not a seal of truth. As always, it runs on trust in the source of the content — not the technical source. And trust has to be built; you can’t decree it or stick it on. I simply trust the Tagesschau more than an anonymous account on the internet. And when in doubt I don’t care whether it was made with AI or prepared some other way so that it’s consumable.

There will never be a statutory label for lies and fraud, no matter how badly we wish for one. But then let’s please not pretend that a sticker reading “AI” does the truth a favor. And in the same breath try to fake AI literacy and trust in the digital chaos into existence.