Last week, Figma added an important button to its interface, labeled “Make designs.” The company announced other AI features, too—like automatically naming layers—but those features were largely overshadowed in community discussions by the idea of generating screens based on a text prompt, and what it might mean for our work.
The response I’ve seen from the design community has been varied, but the top categories I’ve seen are as follows:
- This will increase designers’ efficiency if designers learn more about business. Synergy!
- I can’t wait to play with this and make something really weird.
- This will replace designers; we’re doomed.
- This won’t replace designers, but we’re still doomed because execs don’t understand what designers do.
One thing I’ve noticed about all these responses is the underlying agreement that this feature represents a coming fundamental change to our discipline. I think the real meaning of this development is probably a synthesis of all these responses, which would probably point back to foundational questions we need to tackle in the discipline rather than a new seismic change for which we must prepare.
One such issue represented in these responses is the idea that design has now (just now) been commodified, and as soon as business leadership realizes that, it’s all over. So, it makes sense that the discussion largely turned toward what differentiates human designers in a “make designs” world. One answer I noticed coming up a lot was that designers (and “real” design) are defined by “taste”: a special factor or the secret sauce of being a designer. People posting about AI in design in the past week have suggested that human designers have “taste,” that they are “tastemakers,” that they are defined by “craft,” etc.
But we will need a stronger explanation than this if the AI generation of screens takes hold (more on that later). After all, taste, says sociologist Steven Shapin, is “among the most private, arbitrary, and least-discussable of all subjective modes.” There’s no accounting for it, as the saying goes. And if there’s no accounting for it, we probably shouldn’t be resting our profession on it.
A more robust argument must rest on what design is and does.
The Material Design guidelines say that “label text is the most important element of a button. It describes the action that will occur if a user taps the button.” The phrase “make designs,” I would argue, does not describe what occurs when the button is pressed because design is not merely the production of an image based on patterns. Design, as an object, also does more than just present an image.
As I’ve written before, design is neither an object nor a process. Design is a system; one that collects, synthesizes, and represents inputs from the other systems around it, capturing and transferring meaning. Back then, I also wrote:
...Tools, for example, that place new screens into Figma files based on a statement of intent (a “prompt”), like all generative products, create ... a mass-object: a series of representations of an idea that lack exact origin or subjective meaning. We cannot know the subjective decisions that brought [a generated screen] to us, so we cannot know if it actually “works.” Ironically, at the exact point in which we’ve removed humanity from the production of an object, it loses all meaningful function.
We need to zoom in on the lack of subjective origin: design being a system means that it’s inflected by all the social, economic, and political systems that shape the lives of those who create it. The object itself is just one outcome of that process. The other outcomes belong to users, culture, and the world.
In that essay, I wrote that we would need to essentially inject subjective meaning back into any generated design, making the distance from starting point to usable artifact pretty wide.
As I wrote last year (and have written for the past eight years), generative AI will likely have a place in the interface soon. However, placing it at the beginning of the process could be costly for designers, execs who believe in certain definitions of efficiency, and users.
That’s precisely why I don’t think the wholesale generation of screens is as portentous as it feels right now. I believe it is the visible, somewhat distracting face of the idea that language models (let’s be more specific than “AI”) can integrate into our workflows meaningfully.
Renaming layers, adding descriptions to things, finding and addressing a11y bugs—things that have a tangible, immediate impact on the quality of the final product—will take hold long before a language model can successfully describe an interface that connects with humans if it ever can.
In the meantime, we have time to step back, breathe, and think about the framing and situation of our practice as the people creating the interface.
If we can recognize, acknowledge, deal with, and build intention around all the subjective and systemic influences that lead us to decide how the interface works and adopt tools in full recognition of their limitations, then we’ll become irreplaceable.
