A friend recently redecorated his living room using AI. They fed some photos into a design tool, got back a mood board, colour palette, furniture suggestions. The result looked professionally good, but it felt a bit like listening to a Spotify playlist called Chill Vibes: technically fine and in theme, but never anyone's favourite.

Everyone has the same tools now

The gap between "someone with taste" and "someone with a subscription" nearly closed. Everyone is using the same models trained on the same data, and if impressive is what you're after, congratulations, you're done. The prompt is a thin layer. If the only barrier between you and the output is a prompt, your output is interchangeable with anyone else's.

What curation is actually worth

Curation has always felt like less than creation. The curator didn't make anything, they just chose. But flip the economics: the lower the cost of creation, the higher the value of selection. When anyone can generate a thousand images in an afternoon, the person who picks the right five is doing the harder work.

A travel blogger who visits 30 restaurants and writes about 3. A newsletter curator reading 200 articles and sharing 5. The "just" in "just chose" is doing more heavy lifting than it gets credit for.

The last 15% is the whole point

AI gets you 85% of the way on almost anything. Code, copy, design, photography. That percentage used to be expensive, but now it's table stakes. The interesting question is what happens in the remaining slice, because that's where craftsmanship lives.

The typographer who kerns by eye, the designer who knows which element to remove, the developer who feels when an abstraction has one layer too many. You don't prompt your way to that kind of judgement. You accumulate it over years of paying attention.

We recently had our own website redesigned. Could we have prompted it? Sure, but we hired humans with taste instead, because the difference between a generated layout and a designed one is the same difference as between the living room and a place you call home.

What clients pay for is the judgement that turns competent into right. The flood of competent output makes the crafted version more visibly different, not less. AI is widening the gap between "this works" and "this is exactly what it should be."

Taste is not skill

A skilled photographer can take a technically perfect photo of anything. A photographer with taste knows which moment to capture and which to skip. Skill is trainable: exposure, composition, colour theory. Taste is accumulated. It comes from looking at thousands of photos, developing opinions about which ones matter, and being willing to throw away work that's good but not right.

The same applies to marketing, to writing, to building software products. You can learn the mechanics, which are, if done right, automated anyway. What you can't automate is the willingness to have opinions. To say "not this" when the output is perfectly fine, to choose the harder path because the easier one produces something forgettable.

Ask any songwriter who's thrown 99% of his scribbling in the bin. That's the cost of taste.

What you'll actually get paid for

If AI handles execution, what's left? You'll get paid for knowing what to produce, how to refine it, and when to stop.

That's taste. And it compounds the way AI tools don't. A model doesn't get better because you used it a thousand times, but your taste does. Every project, every decision, every moment you chose "not this" added a data point to your internal filter.

The living room looked fine. The interesting rooms, the ones you actually remember, are full of choices that a model wouldn't make. A weird lamp. A colour scheme that shouldn't work. A photograph from a trip that means something to exactly one person.

AI builds Chill Vibes playlists. Taste builds mixtapes. That's the moat.