Every good design has at least one key detail—an organizing principle that connects to brand strategy while creating hierarchy, guiding attention, and establishing invisible structure. Good designers notice them immediately while most people respond to them subconsciously. These details often emerge through iteration, sometimes requiring only pixel-level adjustments to unlock their function, but without them, even technically proficient design falls flat with arbitrary elements and muddy hierarchy.
AI gives us the ability to render and manage these kinds of details with greater precision than ever before—but only if we understand what it’s actually good for. The fear that AI will replace design thinking misses the point entirely: good design is not generated, but production will be increasingly generative. The distinction matters.
From the Canvas to the Chat
Over the last decade, shared canvas tools like Figma have helped teams increase the rigor of their work by establishing detailed, functional, and distributed design systems. Rather than thinking of the design documentation of a website as just a series of page layouts, we instead consider it a system of rules manifest in tokens, components, and adaptive modules. It takes a lot of labor to construct such a system; little of it is actual design work, but the tedium of typing. System building is still necessary, but it’s activity that I think will largely happen elsewhere. The canvas will remain an essential tool, but the nature of systematic labor – and who does it – is changing. More on that in a moment. But it’s also worth considering another aspect of the canvas, because as much as it has extended the reach of design and the depth of its detail, it has also created problems for more than just designers.
The purpose of the canvas has been to make the collaborative translation of a design system into code and other produced, on-brand artifacts consistent. The canvas served as the source of truth, and as things were built, everyone could refer back to it to ensure the right attention to the right details.
Of course, that also created a process where creativity conflicted with capability. Designs could be approved and then discovered to not have considered all possibilities or to have overlooked a functional detail or even to have depicted something that simply wasn’t possible. Sometimes, that resulted in the canvas being updated to depict what actually ended up being built. But it always resulted in more time and tension than anyone wanted. This sort of thing is to be expected when design work happens entirely in a speculative environment; that’s what the canvas is. Increasingly iterative processes have been explored to reduce this sort of tension, but never fully solved it. I think AI will because it reduces the latency between creativity and construction.
Generative Design Systems
We have already built a system that uses fundamental skills to translate brand standards and accelerate the production of a variety of assets – from websites to ads, emails, and documents.
Our goal was never just to make production faster, but to make it possible for a strategist to work directly with an informed agent on which they can rely to produce on-brand, designer-approved assets fast enough to enable true experimentation and the kinds of multivariate testing that everyone envisions but rarely sees put to action. Up until now, workflows that pass concepts, copy, and assets back and forth among people makes this slow, expensive, and error-prone. Centralizing skills and perspectives within an agent solves that.
On the design side, the agent is trained on documentation that has already been tested and vetted human designers. The difference is in when they do their work – up front and at more predictable and change-determined intervals – and how much unnecessary labor has been taken off their plate. For example, a designer will still create brand systems in Figma. But rather than creating a functional component system that they expect to use for future asset production, they’ll create a style-guide system to train a connected agent. That training will result in brand articulation skills. They’ll also define the foundational skills that maintain best practices across format – those fundamental design concepts that apply to web pages, ads, emails, documents, and the like – in order to train the agent to reliably combine its brand understanding to well-designed structure. The busy work of building and duplicating templates goes away, but the foundational work of establishing the actual elements of design remains. In other words, the true craft remains.
Even the best automated system won’t replace all labor, though. Systems need to be tended to, and the elements of that which are crafted by a designer will be nurtured and maintained by one as well. I have already updated various design skills in our system numerous times. I recently saw a designer write that their day to day work is “more docs than you’d think.” And that speaks to two realities: design is ultimately about creating order, which must then be documented and applied somehow. But it’s also true that though a system can be translated into text, it can rarely be conceived that way. For designers, the canvas will likely remain the most useful and efficient tool for establishing a visual system. But it will not remain the node through which the system is applied.
Design Leadership
The conventional wisdom that design leadership is about rising above details to focus on “big picture” strategy has always felt wrong to me but the AI transition, I think, proves why.
The more we rely upon automation, the more necessary our attention to detail, not less. It requires us to think in advance of the machine and document everything; it requires us to thing after the machine, and review everything. Recognizing and creating key details becomes more crucial, not less.
There is no big picture without detail; the grandest strategic vision fails without countless small decisions made with precision, and the best design leaders never lose their eye for the detail that makes everything work, understanding that their responsibility is seeing details more clearly and helping others see them too. The labor that we hand over to the AI creates space for us to provide this attention.