When I started designing a new comprehensive design system, I expected the main challenge to be designing components, defining rules, and writing documentation. It turned out, however, that the real challenges lay elsewhere. In particular, once AI becomes a significant consumer of such a design system, some of the fundamental assumptions on which today's design systems are built begin to lose their validity. At the same time, entirely new blind spots emerge—areas for which no underlying assumptions have ever been defined.
This article presents five working hypotheses that emerged during the design of a new generation of design system. It does not attempt to propose a finished theory or a new standard. Its primary goal is to open a discussion about whether today's design systems are designed for a world in which humans are no longer their only users—and whether the time has come for a fundamental shift in some of the paradigms we have taken for granted.
Design Systems
Design systems have become a standard part of digital product development over the past few years. Although they differ in scope and quality, most of them are built on a similar assumption: their primary users are people. Designers, developers, testers, and product managers. Today, that assumption is no longer self-evident.
AI is gradually becoming an active participant in product design, implementation, and governance. It is increasingly used to design interfaces, write code, create components, check consistency, and explain the rules of a design system. This leads us to ask a simple question:
Are today's design systems designed for a world in which humans are no longer their only users?
Artificial Intelligence in Design
We often hear the same observation repeated: AI is changing this or that, AI has taken jobs away from one profession or another, AI has destroyed one industry after another. I do not believe that artificial intelligence is destroying anything. It is certainly changing things—but not necessarily in a destructive way.
AI is not a monster. It is simply a new tool. One that can be remarkably useful when we know how to work with it, use it for the tasks it does well, and provide it with data it can understand. In more exact disciplines, such as engineering or software development, this has become fairly obvious, and effective methods for integrating AI into existing workflows have evolved very quickly.
Things become more complicated in less formal and more creative disciplines. AI is already capable of generating large amounts of content, including user interface designs, but much of it is still based on relatively routine and repetitive recombinations of familiar patterns. An experienced eye can often tell that a piece of text, an image, or a UI was generated by AI. The reason, however, is not that AI agents are inherently incapable. Rather, it is that we have not yet learned how to teach them more of the creative process or provide them with better knowledge structures for producing richer and more innovative designs.
AI Reveals Previously Invisible Weaknesses
Working with AI exposes the limitations of the current UX design model. The hesitation and repetitive behaviour we often observe when AI designs user interfaces reveal that design systems have been built on assumptions that are no longer universally valid. They were created primarily for people—as a way for designers to communicate knowledge and outcomes to developers, users, managers, and other stakeholders. Both their content and their form reflect that purpose.
Neither the form nor the content, however, is optimal for AI. And if AI is to become another—perhaps even the primary—consumer of design systems, then not only their form and content may need to change, but possibly the very concept of what a design system is.
Hypothesis 1
A Design System Is No Longer a Component Library. It Becomes a Knowledge Model of the Product
Today, a design system is often equated with a component library accompanied by documentation. Components, however, are only one implementation of a much broader and more complex model.
At first glance, it may seem that the primary value of a design system lies in its precise description of how a button or a dialog should look and behave, and how developers are expected to implement it. But that is only the outer layer. What really matters lies beneath it: knowledge. The knowledge of why a component exists in the first place, when it should be used, which problems it solves, and how it relates to the rest of the product.
If this perspective proves to be correct, we will need to stop thinking of a design system as a catalogue of UI elements and begin to understand it as a formalized model of a product's user interface.
Hypothesis 2
The Real Value of a Design System Lies in Rules, Meanings, and Design Intent
Anyone can create, modify, or generate components. Preserving the reasons why they exist and what purpose they serve is far more difficult. In my view, the design systems of the future will therefore have to be built primarily around rules, constraints, semantic relationships, and design intent. Together, these define what constitutes the right solution in a given context.
From this perspective, components are just an answer. An answer to a question posed by a precisely defined design intent, with rules, constraints, and relationships defining the space in which that answer can exist. Answers can be derived and generated—and this is precisely what AI is exceptionally good at. What it does not know are the questions, the rules, and the constraints. Those are what a design system must be able to preserve and communicate.
Hypothesis 3
Documentation Evolves from a Reference Manual into a Knowledge Architecture
Most design system documentation today is organised as a collection of pages, chapters, and examples. This model serves people well, but it is far from optimal for machine reasoning. AI struggles to extract the underlying principles, intentions, rationale, and the many implicit pieces of knowledge hidden within it. Fortunately, this form of documentation is not the only possible representation of knowledge.
Documentation is no longer the primary artefact of a design system. Instead, the primary artefact becomes the knowledge architecture itself—for example, a network of atomic rules, principles, and relationships from which different views can be generated depending on the user's role, task, or context. Documentation is no longer the source of truth, but merely one possible projection of the same knowledge model.
Hypothesis 4
The Role of the UX Designer Shifts from Designing Screens to Designing Rules
As individual screens are increasingly created with the assistance of AI, the primary value of a UX designer will no longer lie in the ability to design those screens directly. Instead, it will shift towards the ability to formulate rules, define design intent, establish constraints, organise knowledge, and create the framework within which specific designs are generated.
Rather than designing interfaces directly, UX designers will increasingly design the system that makes those interfaces possible.
Hypothesis 5
Dynamically Generated User Interfaces Force Us to Rethink Fundamental Principles
Most of today's UX and information architecture principles were established at a time when the user interface was considered a fixed artefact. But what happens when parts of the interface are generated only at the moment they are needed?
What does consistency mean if every user may see a different version of the same product? Or even if the same user encounters different interfaces in different contexts? What does information architecture mean if navigation becomes just one of many possible paths to information? And how does the very concept of usability change?
I do not have definitive answers to these questions. But I believe they are precisely the questions that will shape the future of UX design and information architecture in the years ahead.
Conclusion
The existence of AI alone would not necessarily justify creating a new generation of design systems. But it certainly raises questions that make us realise we may need one. If the previous hypotheses are correct, then AI gives us a reason to redefine the very nature of a design system.
I do not consider these hypotheses to be definitive conclusions. They are starting points for discussion as well as for practical validation. All of them emerged while designing a new design system that attempts to respond to the changes brought about by AI. My goal is not to prove that they are right at all costs. Rather, I want to verify whether they lead to a design system that is more understandable, more sustainable, and ultimately more usable—for both people and autonomous systems.
It may well turn out that AI itself does not actually require a new generation of design systems. Perhaps it has simply helped us discover that some of the assumptions on which today's design systems are built are no longer sufficient.