The Map of the ObviousThe Map of the Obvious

A client brief usually has one word that matters more than the rest. For Renvia Capital that word was "exclusive". For a clinic it might be "trust", for a startup "innovative". The client writes it as if it explains itself. It doesn't. Somebody has to decide what "exclusive" looks like, and that decision is most of the job.

Generative AI is now being sold as the tool for exactly this step. Write the word, get fifty concepts. I use AI tools in concept work too, so I'm not writing this from outside. But I think they're useful for a different reason than the one in the ads.

01What you get back

An image model is trained on a huge number of existing pictures and the text around them. So when you ask for "trust", it gives you what usually sits next to that word: a shield, a handshake, a dark blue, two hands holding something small. Ask for "exclusive" and you'll mostly get gold, crowns, keys, black backgrounds and serif capitals.

These are fair answers. They're also the most common ones, and producing the most common answer is what these models are built for.

Saussure, the Swiss linguist a lot of semiotics starts from, had a simple idea that helps here. A word means something partly because of all the other words that could have stood in its place and didn't. Later writers called that set of alternatives the paradigm. An image model shows you that set very quickly. It doesn't choose from it, at least not for a reason connected to your client.

02How I use it

For me, the list of obvious answers is the useful part. Collecting the clichés for a concept used to mean a lot of searching and relying on memory. Now I can see most of them in a few minutes, and I know where competitors will probably end up and where the audience has seen it all before. I treat that list as places to avoid, or to go into on purpose if the brand really needs the familiar sign. Sometimes it does.

For Renvia we didn't go toward the gold and crowns. The idea was designing for filtration rather than attraction: the brand should keep the wrong people out, not try to impress everyone. That gave us a strict monochrome palette, an R monogram on a modular grid, and print finishes like blind embossing, laser die-cutting and deep engraving. Blind embossing has no ink. You only see the mark from its shadow, when the paper catches the light. For a closed network, I think that's more honest than gold.

An AI tool could easily make a black R. What it couldn't do was decide that, for this company, exclusive meant filtering. That came from understanding the business.

03Paul Rand and the one logo

There's an old story I think about a lot here. In the mid-1980s Steve Jobs asked Paul Rand to design the NeXT logo and wanted a few options. As Jobs told it later, Rand said no: he would solve the problem, and Jobs would pay him. He came back with one logo and a booklet explaining why.

I don't work like Rand, and most clients today wouldn't accept that. But his point gets stronger now. Options are becoming cheap, and anyone can generate fifty. What clients still need from us is the reasoning behind one.

04Where it goes wrong

The danger is simple. If you take one of the fifty images the model gave you and develop it, the average made the decision and you didn't notice.

I see this risk most with young designers. I've trained and reviewed juniors at the studio for years, and the hard part was never the software. It was explaining why one idea fits a client and another doesn't. A finished-looking image skips that conversation. It looks like an answer before anyone has asked the right question.

So for now my rule is that AI can help at the start of a concept, not at the end. I'm not sure that rule will still hold in a year.