Good enough is the new ugly

AI is making it easier to create decent work. That is exactly why decent work is becoming invisible.

One of the instructions I have repeated the most while editing photos with AI is something like: improve the lighting, make the background look better, fix the hair a little, but please do not make the person look AI-generated.

I have asked for almost the same thing when working on landing pages, articles, videos, presentations, and other creative work. Make it better, but do not make it look like the tool that made it better.

It sounds contradictory, but I think it captures something that is already happening. The problem with AI-generated work is no longer that it always looks bad. A lot of it looks good enough, and good enough is becoming so abundant that it is starting to look bad by default.

To be clear, I use these tools every day. I can now create websites, images, videos, and written content much faster than I could a few years ago. I am not nostalgic about the time when producing a decent landing page required weeks of work, several people, and a budget most early-stage companies did not have.

This is obviously progress. But every time a technology makes a certain level of quality easier to reach, that level eventually stops being impressive.

The floor moved

A few years ago, having a reasonably polished website was already a differentiator for many small companies. The copy was clear, the design looked professional, the mobile version worked, and the whole thing did not feel like a WordPress template from 2012. That alone placed you ahead of a large percentage of the market.

Today, someone with limited design and frontend experience can use AI coding tools to create something decent in a weekend. The website will probably have clean typography, rounded cards, a large headline, a few gradients, testimonials, and some kind of bento grid explaining the product.

The result may genuinely look good. The problem is that thousands of other websites now look good in exactly the same way.

The same is happening with content. AI can write a reasonably clear LinkedIn post, generate a hook, organize the argument, and turn it into five lessons. It can edit a short video with captions, zooms, B-roll, and the pacing that currently performs well on social media. It can create a professional headshot, a podcast cover, or an advertisement that would have required a freelancer before.

Again, none of this is bad. The average quality is clearly improving. But the internet is filling up with work that is technically competent and extremely easy to forget.

This is why I think good enough is becoming the new ugly. Not because the output is objectively worse, but because our perception of quality moves together with the baseline. Once everyone can reach the previous standard, the previous standard stops communicating anything.

What makes something look AI-generated

Today we still associate AI-generated work with obvious mistakes: strange skin, excessive sharpness, meaningless details, plastic faces, weird hands, generic gradients, or writing that uses the same five expressions repeatedly.

Most of those problems will probably get fixed. The models will get better, people will learn how to use them, and the most visible artifacts will disappear.

But I do not think that will eliminate the “AI-generated” feeling. The deeper issue is not a specific visual artifact. It is that a lot of the work feels like the average of everything that already exists.

A landing page looks like what the internet has collectively agreed a modern landing page should look like. A founder post sounds like what thousands of previous founder posts have taught the model a founder should sound like. A video uses the same editing patterns because those patterns have already worked for other videos.

You can see that something is polished, but you cannot see why any particular decision was made.

That is what starts to feel artificial to me. The typography is correct, but it could belong to any company. The article is readable, but you cannot imagine a specific person needing to write it. The video moves constantly, but the editing is not really adding anything to the story.

Everything is reasonable. There is just no particular point of view behind it.

This also explains why asking AI to “make it more professional” often makes the result worse. Professional is an extremely vague direction. Without more context, the model has to move toward the center of the category, which usually means removing whatever was unusual and replacing it with something safer.

At some point, professional becomes another word for familiar.

Taste matters more when execution becomes cheap

I recently said that great editors and people with taste are immune to AI. Immune is probably too strong, because their jobs will clearly change too. But I still think their value is much more defensible than that of someone whose main skill is reproducing an existing style.

If your value as an editor is knowing how to imitate the captions, transitions, sound effects, and pacing of the videos that are currently popular, AI will become very good at that. It is a recognizable pattern, there is a large amount of reference material, and the result can be evaluated relatively quickly.

But that is different from understanding why a specific story should be slow, where silence is more effective than music, which moment deserves attention, and which technically good scene should be removed because it weakens the rest.

Taste is not only the ability to generate something attractive. It is the ability to decide what belongs and what does not.

That sounds obvious, but it is much harder than execution. An AI tool can generate 30 decent versions of a landing page. The difficult part is deciding which direction reflects the company, which one the customer will understand, and which one will still feel relevant after the current design trend dies.

The same applies to writing. Producing more words is not the problem anymore. The problem is having something worth saying, noticing what is generic, and deleting the paragraphs that sound intelligent but do not actually add anything.

In my case, AI is very useful when I already have an observation or an argument. It can help me organize it, question it, identify missing pieces, and improve the language. The results are much worse when I begin with nothing and ask it to come up with something interesting.

It can produce an article, of course. It just tends to produce the article that someone could have written.

More content does not necessarily mean a stronger brand

The natural reaction to cheaper production is to produce more. If a company can now create ten videos instead of one, publish every day, test dozens of advertisements, and redesign its landing page every week, why would it not do it?

There are good reasons to take advantage of that leverage. More iterations can help you learn faster, especially when you have little data and are still trying to understand what resonates.

But there is also a real risk. A company can dramatically increase the amount of content it publishes while making its identity weaker.

One post sounds like a venture capitalist. The next one sounds like a motivational creator. The website looks like an AI startup, the sales deck looks like McKinsey, and the videos copy whatever editing style was popular that month. Each piece can look fine individually, but together they do not form anything recognizable.

A brand is partly the consistency of the decisions it makes. What it talks about, what it ignores, how it explains things, what it finds funny, what it refuses to publish, and how it wants people to feel.

Before generative AI, some of that consistency came from limitations. The same designer created everything. The founder wrote most of the posts. The company reused formats because producing new ones was expensive.

Now the range of possible output is almost infinite. That sounds like complete creative freedom, but it also means that companies need much clearer constraints. Otherwise, every new asset becomes a remix of a different part of the internet.

This is one of the reasons I think creative direction will become more important. Someone still has to define what the company should look and sound like, and more importantly, why.

Originality starts before you open the tool

There is a lot of discussion about prompt engineering and how to make AI generate more original work. I am sure prompting can improve the result, but I think most of the originality has to exist before the prompt.

It comes from the experiences, references, conversations, and opinions you bring to the tool.

If everyone is consuming the same startup podcasts, following the same accounts, reading the same newsletters, and using the same successful companies as references, the output will converge even with better prompts.

The more interesting inputs usually come from somewhere else. A conversation that was never published. Something you observed while working with a customer. A failure that contradicted the standard advice. A reference from architecture, film, music, science, or an industry unrelated to the one in which you are working.

This is also why I do not think AI automatically gives someone taste. Taste is developed through exposure, practice, criticism, and making a lot of decisions that turn out to be wrong. You learn that something looks good, then you try to understand why. You copy it, realize it does not work in another context, and gradually become better at separating the principle from the surface.

AI can accelerate that process. It can help someone explore references, generate variations, and obtain feedback much faster. But the person still needs to pay attention.

There is no prompt that replaces having an opinion.

How I am trying to use it

I am still figuring this out, but there are a few principles I am trying to follow.

First, I try to begin with an observation instead of asking the model to find one for me. The initial material can be messy. It might be a voice note, a conversation, a paragraph, or something I have been thinking about for a few days. But there has to be something specific underneath the output.

Second, I use AI to generate range, not to make the final decision. It is useful for showing me alternatives I would not have considered, but I do not assume that the most polished answer is the correct one. Fluency makes average ideas feel more complete than they actually are.

Third, I try to use real references and explain what I like about them. Saying “make it look like Apple” is almost useless. Is it the spacing, restraint, photography, typography, product focus, or the fact that Apple can remove information because people already know the brand? Copying the surface without understanding the context usually creates a bad imitation.

Finally, I edit aggressively. I remove phrases I would never say, replace hypothetical examples with real ones, combine sections, and delete anything that seems to exist only because articles are supposed to have it. The goal is not to pretend AI was not involved. I do not care about that. The goal is for the final result to still reflect a real person making decisions.