AI production · [3] min read

We were drawing long before we were writing. A bison on a cave wall, a stick figure on a toilet door, a small man in the IKEA booklet holding the wrong screw with complete confidence. Visual communication is the oldest interface we have. It just keeps changing its outfit.

Many forms, one job

A floor plan is a visual argument about how people will move. The safety card on a plane shows a remarkably calm person sliding into the sea. A traffic light gets three colours and no adjectives. A chart in a board meeting is usually a decision wearing a costume.

The forms change. The job doesn’t. Get something out of one head and into another, with as little lost on the way as possible.

Then the pictures got cheap

AI made images almost free. Type a sentence, get a photoreal anything. Lamps on marble, lawyers in golden hour, a croissant on the moon.

The bottleneck moved. It used to be “can we make it?” Now it’s “do we actually know what we’re making?” The machine is brilliant at answers. It is not great at questions. It does exactly what you said, not what you meant, which puts it in the same category as every intern and most clients.

A prompt is a brief. A vague brief used to produce one mediocre layout and a polite meeting. Now it produces two hundred mediocre layouts by lunch, all in 4K, several with six fingers.

A small experiment, with a pigeon

This week a job ad asked me: “Are you up to date with new tools?” Fair question. I could have typed “yes”. Instead I turned my CV into a 20-page manga, using a Higgsfield feature that had been released two days earlier. There is a deadpan pigeon in it. He has opinions about atriums.

The drawing took minutes. Everything around it took two days.

The story had to fit twenty pages and still be a CV. The sidekick was redesigned three times: a floating text cursor, then a chalkboard, then a plain black tablet, because some ideas only work in your head. Eleven locations each needed one exact name, because the model doesn’t know the balcony on page 1 is the same balcony on page 18 unless you tell it, politely and consistently. And the script file was rejected twice before the tool agreed to read it.

None of that is drawing. All of it is visual communication.

The work moved upstairs

AI didn’t remove the work. It moved it upstairs, out of the hands and into the decisions.

Less rendering, more deciding. Knowing what to cut. Keeping a world consistent across a hundred images. Writing a brief a machine can’t misread. Noticing that “IT’S PERFECT!” means there are six fingers on that hand.

None of these skills are new. Editors have them. Art directors have them. Architects definitely have them. I trained as one, and buildings teach you early that nobody admires the beam. They admire the room the beam makes possible. Skip the beam and the room still looks great, for about four seconds.

Up to date with which tools?

So yes, I’m up to date with new tools. This one was two days old.

But the more useful question might be the reverse: are we still up to date with the old ones? Structure, story, consistency, the ability to say no to a very pretty wrong picture. The tools will be replaced by next Tuesday. Those won’t.

Structure for ideas. Ideas with structure. Even when the ideas arrive in 4K.