Why I Do Not Let AI Create or Modify Styles

AI can generate styles, but the problem is that it does not really understand what it is doing. Many times it even spends minutes thinking through changes that would take seconds manually. Adjusting spacing, correcting alignment, or modifying visual hierarchy is usually faster to do directly than waiting for AI to interpret what should happen. The result also rarely ends up being exactly what you wanted.

AI does not understand aesthetics. It does not understand visual judgment, visual sensitivity, or why something works visually. It can repeat common patterns because it has seen thousands of similar examples, but that does not mean it understands the decisions behind them. Many times it changes small details that were already correct and ends up altering the entire visual result without realizing it.

It also does not understand art. Visual work is not only logic or a mathematical combination of CSS properties. Some decisions exist because of balance, intention, contrast, rhythm, or composition, and AI simply does not perceive any of that. For it, styles are statistical patterns and probable combinations of classes. It can visually approximate something, but it does not really understand why that thing exists or what it is trying to communicate.

The problem becomes even more obvious when it tries to modify something that already exists. AI does not understand intention. It does not know why certain decisions were made or which parts were deliberate. It starts changing spacing, sizes, alignments, or visual hierarchies without real context. And many times those changes seem small until they slowly start breaking the overall consistency of the project.

With Tailwind this becomes much more noticeable. AI seems to prefer using Tailwind in the most fragile way possible. It mixes different patterns, adds unnecessary classes, and ends up producing structures that are difficult to maintain. Many times it does not even respect how the project is already organized and simply keeps adding classes on top of what exists as if the only goal were to make something “work” visually in that moment.

Even worse: it frequently does not understand Tailwind versions. It mixes Tailwind v3 concepts inside v4 projects, uses incompatible configuration patterns, and ends up suggesting nonexistent or outdated APIs. Sometimes it even invents complete solutions that do not exist inside the framework. The problem is that, superficially, many of those answers look convincing until you start reviewing what it is actually doing.

And that is probably the most dangerous problem: at first everything looks like it works. The styles compile, the component renders, and visually something appears on screen. But over time the project starts losing consistency, maintaining styles becomes harder, and every modification adds more visual and technical noise. What looked like acceleration ends up becoming more time spent correcting, cleaning, and reorganizing things that should never have been left that way in the first place.

That is why I do not let AI create or modify styles. Not because it cannot write CSS or Tailwind. It clearly can. The problem is that it still does not understand aesthetics, art, or intention, and that ends up showing directly in the final result.