Every poster looks the same now

I read an article on RMC Conso yesterday and it put a number on something I had been feeling for months without being able to name it.

Carrefour went from 37% AI-generated images in its promotional catalogues at the end of 2025 to 100% since the beginning of 2026. Not a pilot, not a corner of the catalogue. All of it. Auchan is running six generative models with a partner and has extended the test past its garden catalogue. Intermarché is now printing AI posters and hanging them in the aisles, next to the bread. A journalist photographed one on rue Lecourbe in Paris.

None of this is a scandal. It is a spreadsheet. Carrefour stopped printing paper prospectus in April 2025, Auchan followed in January 2026, and the images still had to come from somewhere. Production costs for a full catalogue were cut in half. Auchan reports a 19% higher click-through rate and a 7% lower cost per click. Those are good numbers. Anyone running a business would look at them and do the same thing.

And yet.

The tell

Here is the part I cannot unsee. Olivier Dauvers, who has been watching French retail for decades, put it better than I could:

AI, however intelligent it is, makes something pretty and clean, but nothing particular to the brand, nothing that carries its graphic identity beyond the logo. The result is that whatever the subject and whatever the retailer, every poster made with AI looks the same. You can spot them from three kilometres away.

That last line is the whole problem. You can spot them from three kilometres away. I do it now without trying, and once you start you cannot stop. The same soft studio light. The same shallow depth of field. The same impossibly glossy tomato. The same slightly wrong hands. Three different chains, three different price positions, three different customer bases, one visual language.

It is not just supermarkets. It is the poster in the boulangerie window, the flyer taped to the school door, the Etsy listing photo where the product is clearly a render and the seller clearly hopes you will not zoom in. Same texture, same lighting, same nobody-in-particular.

Brand identity is, by definition, difference. A generative model is, by construction, a machine that converges on the average of what it was trained on. Asking it to produce identity is asking a device optimised for the middle to give you an edge. It will hand you something beautiful and generic, and generic is the one thing a brand cannot afford, because generic is exactly what your competitor also received this morning.

What actually gets lost

The obvious answer is jobs, and it is not wrong, but it is not the part that worries me most.

Carrefour says its teams saved more than fifteen days, now spent learning to write better prompts for Gemini and auditing image quality. I believe them. That is a real, honest answer, and it is more thoughtful than most. But look at what it describes: people who used to make images now supervise a machine that makes images. The skill being cultivated is the skill of asking. The skill being retired is the skill of seeing.

That matters because of where the pipeline starts. The ugly local supermarket poster, the one with three fonts and a clip art tomato, was somebody's first job. It was bad, and it was the first rung of a ladder. You made the ugly poster, someone told you why it was ugly, and five years later you were making good ones. Remove that rung and the ladder still has a top, for a while, staffed by people who climbed it before it disappeared.

The second thing we lose is stranger and harder to argue in a meeting: the human error. The typo on the chalkboard. The photo taken with a phone in bad light where you can see the actual bread that this actual shop actually baked this morning. Those things were never optimised. That is precisely why they read as true.

The strongest counter-argument

I want to be fair, because there is a real one, and it appeared right in the comments under Dauvers' post: most local retail design was already bad. Genuinely bad. If AI raises the floor from Comic Sans and a stretched JPEG to something clean and readable, that is a win for the person shopping, and pretending otherwise is nostalgia.

I think that is correct and I think it is incomplete. Raising the floor and lowering the ceiling are happening in the same motion. When everything reaches the same acceptable middle, acceptable becomes the new invisible. The 19% click-through improvement is real today, when AI images are still novel enough to catch the eye. Ask again in eighteen months, when every image in the feed has the same lighting. The advantage was never the AI, it was being early.

Where I actually stand

I need to be honest about my own position here, because I am not a bystander.

I use AI every day. It writes tests with me, it reviews my code, it drafts things I then rewrite. I ship a product built by a small team that could not exist at this speed without those tools. I am not going to stand here and tell an overworked shop owner that they should have paid a photographer.

So the line I draw is not about the tool. It is about who is holding the point of view.

AI inside a point of view is fine, and often great. A brand with a real visual identity, a real palette, a real way of framing its world, using a model to execute that identity faster, is using a tool. That is what Dauvers is asking for when he says feed it the graphic charter first. A brand with no point of view, using a model to supply one, is not using a tool. It is outsourcing the only thing that made it distinguishable from the shop across the street.

The tell is simple: could you cover the logo and still know who made this? For almost every AI poster I have seen this year, the answer is no.

Why this one gets under my skin

Probably because of what I do when I am not writing code.

I cut leather by hand. Every piece I make under À la Française and every guitar strap I make under BrudStraps carries evidence of a specific afternoon: a stitch that sits half a millimetre off, an edge burnished a little darker because the room was humid. Those are not defects I have failed to eliminate. They are the proof that a person was there. A machine-perfect object is worth less to me than a hand-made one with a visible decision in it, and I do not think that is sentimentality. I think it is the entire reason anyone buys hand-made anything.

The supermarket poster is a small thing. Nobody frames it. But it is the wallpaper of ordinary life, and we are in the middle of replacing all of it with the statistical average of everything that came before. Efficiently. At half the cost. With better click-through rates.

If everything is generated from the same middle, the only thing left that carries any signal is the thing that obviously came from a person. That is going to be worth more, not less, and I intend to keep making it.