Rethinking Creativity With AI

Making has always been automated. AI turns the speed up, and speed only multiplies what you bring to it: an opinion, a clear picture of good, and the patience to keep betting on it.

Every few years, someone announces that making things has been automated.

Desktop publishing was going to end typesetting. Photoshop filters were going to end illustration. Bootstrap was going to end web design. Then came templates, UI kits, Dribbble shots, component libraries and design systems, each one taking a layer of the work and making it nearly free.

And yet here we are, still arguing about spacing.

AI is not a new story. It is the same story with the speed turned all the way up. What changed is not whether making is automated. It is how fast you can go from an idea to something you can look at, judge and throw away.

That speed is either the best thing that has happened to creative work, or a very efficient way to produce more of the same. Which one you get depends almost entirely on what you bring to it.

It Was Always Automated

Every generation of tools did the same thing: it lowered the effort it takes to make a decent artifact. Not a great one. A decent one.

Paste-up became desktop publishing. Hand-coded tables became Bootstrap. Bootstrap became component libraries. Component libraries became design systems. And now design systems are becoming a prompt.

Each step raised the floor. It became easier and easier to make something that looks fine. Which is lovely, until everyone's fine starts looking identical. Forty landing pages, one hero section, and every single one of them wants to supercharge your workflow.

But notice what no step did. None of them told you what to make. None of them knew why one version was better than another. The tools automated execution. They never automated judgment, and they never automated taste.

AI is the steepest step on that staircase. But it is still a step.

Open-Ended Means Average

Generative AI is wonderfully open-ended. Ask it for anything and it will give you something. That is also the problem.

A model is, at its core, a very well read guess about what usually comes next. Ask it for a landing page without an opinion and you will get the landing page. Centered hero. Three feature cards. A headline that says "Build faster." A gradient that has seen things.

This is why so much of the vibe coded internet looks like it was made by the same person on the same afternoon. It is the Bootstrap era again, except this time it ships in a weekend instead of a month. Different founders. Same purple.

The average is not a flaw in the model. It is the model doing its job. Without direction, the most probable answer wins, and the most probable answer is, by definition, the one everyone else is getting too.

An opinion is what moves the curve. Not a longer prompt. An opinion.

Have An Opinion Before You Have A Prompt

Opinionated design sounds like a personality trait. It is closer to a document.

It means deciding, in advance and in writing, what your work does and does not do. Not mood board adjectives like clean, modern and minimal. Everyone is clean, modern and minimal. Even the ones that very clearly are not. Real rules. The kind you can check.

This site runs on a handful of them:

• One type ladder. No size exists outside it.
• One body weight. Emphasis comes from ink, never from bold.
• No arrow glyphs. Anywhere.
• Blur is only ever progressive. A uniform blur reads as generated.
• Sound only when someone acts. Never from a timer.

None of these are universal truths. Plenty of great sites break every one of them. That is not the point. The point is that every time AI hands me something, I have a fixed shape to hold it against. The output either fits the opinion or it does not, and I do not have to rediscover my taste every single time.

Without that document, every prompt is a negotiation with the average. With it, AI stops being the author and becomes a very fast pair of hands.

Accelerate The Experiment, Not The Decision

Here is where AI is genuinely, almost unfairly good: trying things.

Creative work has always been a numbers game. Nobody gets the good idea first. You get it on the eleventh attempt, after ten that taught you what you actually wanted. The problem was never the number. It was the price. Each attempt used to cost an afternoon, so most of us stopped at the second, called it exploration, and shipped the first one with better spacing.

Now an attempt costs minutes. When I was working on the hero of this site, I did not debate the entrance in my head. I had several versions running in the real page by the evening. A rotated word drop. A whole headline zoom. A lens with handwritten margin notes. Two of them sounded fine in theory and felt wrong the moment they moved. The lens won, and I only knew that because I could see all of them.

But look at what did not change: the number of decisions. There is still one hero. Somebody still has to pick.

That is the split worth protecting. Let AI multiply the attempts. Keep the decision for yourself. The moment AI starts making the call too, you are back at the average, just faster.

Know What Bad Looks Like

Cheap attempts are only useful if you can tell them apart.

Good is hard to describe. Bad is surprisingly easy to name, and naming it is half of taste. Labels that touch. A heading that leaves one word alone on its last line. Motion that is busy instead of meaningful. Greys that look elegant and fail contrast. A card that exists because the template had a card.

If you can name bad quickly, you can reject quickly, and fast rejection is what turns a hundred cheap attempts into a short list instead of a pile.

So build the vocabulary. Look at things that are excellent and ask what exactly makes them excellent. Look at things that are mediocre and ask what exactly gives them away. The second list is often more useful than the first.

Taste is not a mysterious gift. It is a very long list of things you have learned not to accept.

Bet On Good, Then Compound

This is the part I think most people miss.

When you generate without an opinion, each output is a fresh roll of the dice. Some are better, some are worse, and over time they cancel each other out. You end up roughly where you started, with a lot more files and a Downloads folder that now needs therapy.

When you generate with an opinion, each good decision becomes the ground for the next one. The type ladder makes the spacing easier. The spacing makes the motion calmer. The motion makes the sound feel earned. Nothing here came from one great prompt. It came from a long series of small bets placed in the same direction.

That is compounding, and like any compounding, the early part is boring. The line barely moves. For a while it looks almost exactly like the random one. Most people quit right there, decide AI only makes average things, and go back to generating.

The ones who keep betting on the same idea of good are the ones whose line bends.

Could Anyone Have Made This?

AI did not change what creativity is. It changed the price of trying, and in doing so it removed every excuse for not knowing what you want.

So before I ship anything AI touched, I ask one question: could anyone with the same prompt have made this?

If the answer is yes, it is not mine yet. It is the average wearing my name.

If the answer is no, it is usually because somewhere along the way I said no to something it offered, and then kept saying no in the same direction.

That is the whole method. Have an opinion. Use AI to try more than you ever could. Know what bad looks like. Bet on good, and keep betting.

Something beautiful does not come out of one prompt. It comes out of a direction.