Skip to content

ResearchNote

The return of ornament: design when agents do the building

Ornament did not disappear because it was ugly. It disappeared because it cost too much. As the cost of making things approaches zero, the labor-intensive beauty of the past becomes usable again, and what remains for people is to choose, to hold back, and to check.

  • Design
  • Agents
  • Interfaces

For about a century, design has moved toward less ornament. Carved details on buildings, decorated borders in books, hand-drawn lettering, illustrations made for a single page: one by one they faded, and flat surfaces, orderly type, and generous white space became the default.

We think that movement is starting to reverse. AI agents and image models are bringing the cost of making things close to zero.

Why ornament disappeared

Ornament did not disappear because it was ugly. It disappeared because it was expensive.

  • Anything carved, painted, or assembled by craftsmen costs more the more of it you make. In an age of machine production, simple forms were cheaper and faster to make. Early in the twentieth century, the architect Adolf Loos argued in his lecture “Ornament and Crime” that ornament was a waste of craftsmen’s labor and of material.
  • The same thing happened on screens. Screens come in every size, and digital work is revised constantly, so elaborate illustration and decoration had to be redrawn every time. When smartphone interfaces abandoned realistic textures for flat design in the 2010s, it was not only a matter of taste. Flat forms were easier to adapt to many screens, and easier to change.

Minimalism was a matter of taste, but it was also a response to the limits of the maker’s labor.

The cost of making falls away

Agents today can draw a different illustration for every page, construct letterforms one by one, and rework a layout for every screen size. A round of feedback can turn into several new options overnight. Ways of working once reserved for big budgets and a handful of luxury products are coming within reach of small companies and individuals.

What comes back

When labor stops being the limit, labor-intensive beauty comes back: botanical borders, round frames, lines that look hand-drawn, pictures made for a single page. The richness of posters and book covers from a hundred years ago becomes usable again, not as nostalgia but as a living language.

Distinctiveness comes back too. To save effort, the sites and apps of many companies converged on the same look. As making things gets cheaper, every company, and every person, can afford a look that belongs to a world of their own.

What remains for people

None of this turns out well on its own. As the labor goes away, other things get harder.

  • Choosing. Ask for something “in the style of” someone, and everyone who asks gets roughly the same picture. A style becomes yours only when you decide what to take from it and what to leave out, and turn that into rules of your own.
  • Holding back. When ornament costs nothing, overdoing it becomes the easiest mistake to make. Deciding where to be bold and where to stay quiet is a human judgment.
  • Checking. A generated image can look beautiful at a small size and still fall apart at full size: hands, faces, the shapes of objects, the quality of a line. As making takes less effort, looking closely becomes the center of the work.
  • Writing the rules. What you choose has to be written down as rules that people and agents read the same way: a design system. With those rules in place, agents can make hundreds of screens that still belong to one world.

Open questions

  • When everyone can use ornament, will it become ordinary again and lose its meaning?
  • Can a picture no one drew by hand still carry the warmth of a human hand?
  • When we borrow from past styles, where does learning end and imitating living artists begin?

Related

Version history

  1. v0.3Oct 6, 2026Rewrote the note from a broader perspective, around the idea that ornament is coming back.
  2. v0.2Oct 6, 2026Added the design-token example from this site.
  3. v0.1Sep 24, 2026First version.

This document’s URL will not change. When citing it, please include the version number.

All research