For a long time, making the thing was the hard part.
You needed to know Photoshop.
You needed a camera.
You needed to write.
You needed to edit video.
You needed somebody who could design.
You needed time.
Now you can type a sentence into a box and get fifty logos, twenty images, ten scripts, six landing pages and a video of a fake person walking through a fake hotel lobby before lunch.
So production is not really the flex anymore.
Choosing is.
Taste is basically deciding what survives.
Most generated content is technically fine
That is the weird problem.
AI is already very good at producing things that are acceptable.
The image is clean.
The lighting works.
The copy is grammatical.
The website has rounded cards.
The video has captions.
Nothing is obviously broken.
And somehow it still feels like absolutely nothing.
This is probably why so much AI content feels worse than bad human content.
Bad human content can at least be strange.
Generated content tends to drift toward the middle unless somebody pulls it somewhere specific.
You get the average startup landing page.
The average luxury apartment.
The average cyberpunk city.
The average motivational caption.
The average attractive person looking thoughtfully out a window.
It is competent mush.
The useful skill moved one step later
If generating ten options takes thirty seconds, the valuable part is no longer making option eleven.
It is knowing that nine of the first ten suck.
That sounds obvious until you watch people use these tools.
They generate something.
It appears.
They post it.
No pause.
No editing.
No question about whether it actually fits anything.
The machine produced an answer, so the answer becomes the content.
That is backwards.
Generation should probably be treated more like digging through a pile.
Most of it gets thrown away.
One thing is almost right.
You fix it.
Maybe another thing gives you an idea that is better than what the model actually made.
The output is raw material.
Not a command.
Taste is partly knowing what feels off
This is annoyingly difficult to teach.
Sometimes an image is technically excellent and still looks cheap.
Sometimes a website has every fashionable design element and still feels like somebody copied the internet's idea of "premium."
Sometimes one badly chosen font ruins an otherwise good page.
Sometimes the generated person is too attractive.
The room is too clean.
The shadows are too cinematic.
The sentence explains one thing too much.
The logo is trying too hard.
There is no error message for any of this.
You just look at it and think:
No.
That instinct becomes much more valuable when everybody has access to roughly the same production tools.
More output can actually make this harder
People assume more options make creative work easier.
Sometimes they do.
Sometimes fifty options just give you fifty opportunities to choose the wrong one.
Before generative tools, there was a natural limit.
You might sketch three ideas because making thirty would take all day.
Now thirty is nothing.
So people keep generating.
And generating.
And generating.
Eventually they lose track of what they were trying to make in the first place.
You see this constantly with AI logos and image generation.
Version one had something interesting.
Then there are twenty-seven revisions.
Everything gets shinier.
More detailed.
More "professional."
By the end it looks like an energy drink company designed by a GPU.
The ability to stop is taste too.
References matter more than prompts
People became obsessed with prompts because prompts looked like the new technical skill.
There is definitely skill involved.
But the better question is usually: what are you looking at?
What do you actually like?
What have you saved?
What visual references are stuck in your head?
What internet subcultures have you spent too much time inside?
What brands, films, websites, album covers, stupid forum graphics, old ads, game menus or magazine layouts shaped your eye?
AI can remix what you point at.
It cannot give you ten years of noticing things.
That part still belongs to you.
If two people have access to the same model and one has spent years paying attention while the other has not, they do not really have the same tool.
One has a much better filter.
This will probably make weird people more valuable
Not weird for the sake of weird.
People with specific taste.
People who like things that are difficult to average.
People with oddly narrow references.
People who know exactly why one image works and another one looks like shit.
If everyone can produce polished content, polished content becomes less interesting.
The edge moves toward selection, editing, restraint and point of view.
Basically all the stuff that does not show up in the generation button.
AI made it much easier to make more.
It did not make it easier to know what deserves to exist.
That might end up being the whole game.