I find it exciting to see design systems empower design teams to scale and consistently produce solid products, but I know we’re only scratching the surface of our potential. There’s so much more we can accomplish.
At Airbnb, we’ve been pondering how we might push our design system in new directions, and we’re inspired by design-forward companies that share our desire to craft the future of design systems.
In this chapter, I want to introduce you to a new way of thinking about design systems. Design systems can transcend the walls of a single company to exist as shared standards and customizable tooling with help from the open source community, which can accelerate development and eliminate the need to start systems from scratch. If we’re really bold, we could be creating adaptive, intelligent systems that are context-aware and compose themselves — reducing our workload and ultimately unlocking AI-powered design.
But, I’m getting ahead of myself. Let’s start with nuts and bolts.
Building a common foundation
A designer’s familiarity with the concept of a design system is based on the systems they’ve encountered, what platforms they’ve been tasked to support, and where they’ve worked. Books like this one help us converge on a high-level definition for design systems, but a more rigorous definition will ensure the utility and flexibility of our work.
Finding a standard that supports shared goals will involve decoupling a design system from its implementation, cataloging common UI and associated states, and more strictly defining design primitives and components. This could then be expressed in a file format that’s able to define a component or design system more completely.
An example of differing goals
Currently, existing systems reflect the specific needs of the companies that created them. Since each company is building an entirely independent system, design system development starts from scratch — possibly with help from a web toolkit like Bootstrap, relying on the internal knowledge of the team, and only focusing on top-level needs. As a result, even the best systems contain flaws and lack the necessary tooling to speed development and track results. And if a company’s priorities shift, its design system must shift, causing another section of the design system to be built in the same limited way.
For example, one of the reasons Airbnb created DLS was to minimize and sync differences in UI between our supported Android, iOS, and web platforms. In The Way We Build, VP of Design Alex Schleifer writes, “Universal and Unified define the system’s approach we apply when defining patterns. Is it part of a greater whole? Does it work across devices?”
In our idealized vision, a mockup easily ports between platforms — creating a better design and development experience. This cross-platform UI would give Airbnb guests and hosts alike the same end-product experience as if they jumped between mobile app and desktop web.
In contrast, supporting multiple device platforms was initially less of a concern at Etsy, where the main priority was to scale its web platform. During my time there, the team built the web toolkit with the core website as its primary focus. Later, Etsy expanded upon its toolkit to support different branding elements for other internal web initiatives. Karyn Campbell describes what it was like modifying the Etsy design system while making Etsy Studio. “While we made a conscious decision to depart in some instances with the etsy.com UI in order to birth this new brand, we also retained many underlying components that our design systems team had created.”
A priority at Airbnb was having the same functional and visual voice across platforms. A priority at Etsy was to support multiple web products with varying brand initiatives. Both were valid needs. A shared standard for design systems would need to ensure a solid foundation so that both these and other real-world priorities could more easily be achieved.
Gathering examples of different design system priorities will help create a checklist to make sure your design system standards address real concerns. Any company adopting these suggested standards could be assured their design system development focuses on immediate company needs, as well as adhering to standards they plug in to a growing body of open source code and tools that support most operational transformations that could be encountered.
But what should be in the design system standard?
Imagine a tool that can specify which design primitives, (e.g., fonts, spacing, color — more on that below), components (and their states), platforms, and what documentation and testing are needed to have a fully formed design system. The tool would also allow the designer to specify which components were not yet needed and which platforms could be added later. With this tool, a designer would have a framework stating what aspects of the design system were completed or outstanding. A product manager could export documentation, and a developer could easily export UI and UI tests — no longer needing to translate UI from Sketch to code, or from web code to native implementation.
If created today, not only would this tool provide industry-wide savings, but it would start to standardize the low-level definition of a design system. Working backward, let’s now imagine what kind of definitions such a tool would need in order to exist.
First, decouple the design system from any specific implementation. We’re not creating React components (nor other web implementations), nor Android UI, iOS UI, or even Sketch files. Instead, our system is an abstraction that can be deployed to any target implementation. We’re going to need a file format to describe this abstracted design system. The exported format could be rendered into views by open source modules specific to each target implementation.
Next, codify the definitions of design primitives and components so they are fully expressed in the design system format. Dang, .dsf extension is already in use! Guess we’ll have to settle for the .dang file extension!
Design primitives are the building blocks of a UI. These include specific predefined colors, fonts, spacings, and more. They are foundational visual elements that can be combined into components. Changing primitives is echoed throughout a given design system’s components, and doing so changes the overall feeling of a brand. Additionally, what are components? We’ll also need to codify those. Components are mostly views composed of design primitives and smaller components whose minimal internal logic is mapped exclusively to state and state change. Benjamin Wilkins, Design Lead on the Design Tools team at Airbnb, describes the difference between primitives and components in minute 7:00 of his talk, Thinking in Symbols for Universal Design.
Next, we’ll need to catalog all of the common UI components in use today. Just as a typeface may have its unique take on the letter “A” (the letter’s visual appearance may vary between typefaces while its meaning is maintained), a .dang file would have a text input component that varies in visual representation but not functionality. The catalog will need to group components with their accompanying states (selected, focus, on-tap, error, etc.) and detail interactions to distinguish between mobile, desktop, and TV UI.
What are the benefits of this catalog? To start, functional tests for common components could be easily automated through contributions from the open source community. In many cases, UI engineers would no longer need to write their own tests. The cataloged components would also enable a marketplace of boilerplate design systems that can be installed interchangeably, and against which custom UI can be built and substituted. This means bootstrapping the creation of every design system is no longer necessary.
Lastly, we need to allow for the evolution, growth, and extensibility of design systems built upon the shared standard. Just because we’re aware of which components are needed today doesn’t mean that we’re able to predict all the elements needed for future innovation. A process for modifying existing components or creating wholly new ones is in order. Thoughtfully standardizing our collective knowledge will produce a more consistent user experience, accelerate development, decrease investment needed from individual companies, and enable open source and collective development of next-generation design tools that conform to shared conventions.
Creating a single source of truth
The elements that make up a design system — principles, UI components, patterns, and documentation — create the human-computer interaction layer for our apps. Product designers and system designers are directly responsible for this layer, and therefore should own the design system and its representation in the codebase.
There are 2 hurdles to achieving a single source of truth. First, our current design tools are inadequate. Most only allow us to produce images of UI and prevent designers from achieving product level fidelity. Second, if the implementation of a design system is spread across multiple repositories (Android, iOS, React Native, React, etc.), collected in a Sketch file, and documented on a website, then there really is no single codebase to represent a truthful account of the system. Lacking a single source of truth, the design system — spread out over multiple codebases — becomes an amalgam of sources that easily fall out of sync.
Designers use tools like Sketch, Illustrator, or Photoshop to draw pictures of UIs, yet these are actually just representations of interactive components that look different, behave differently, and contain different data depending upon the state of the app at a given time. As Colm Tuite notes in the article Design tools are running out of track. Here’s how we can fix them, “Think of the number of simple interactions which are commonplace in almost all of our products yet cannot be communicated through our design tools.”
Tuite then mentions interactions and states such as hovering over a button, focusing an input, checking a checkbox, and identifying scroll areas. He points out that our design tools aren’t prompting designers to think with product level fidelity, and so a designer’s work is usually missing some of its most important details.
Sketches of apps are then handed off to developers who have to translate them into working UI. Between the designer and the 4 developers that it takes to convert the design into Android, iOS, React Native, and React, it takes 5 different members of the team begin to bring the design up to product-level fidelity. Since the original sketch was missing details about state and interactions, a back-and-forth conversation between designer and multiple developers is needed to make the designs production-ready. And because the implementation is coded by 4 different humans, it’s likely that unwanted variation creeps into each implementation.
For similar reasons, many designers have focused on sharpening their coding skills for at least 1 platform. There are many advantages to this, but if you’re a systems designer creating components for cross-platform use, coding those components for a single implementation is not yet enough.
The attempt to reach a single source of truth is further complicated when working on cross-platform design systems. Jon Gold identifies places where Airbnb’s DLS workflow could be improved in Painting with Code.
At Airbnb, Gold has taken some exciting first steps toward solving this problem with his project React Sketch.app. With this tool, layered Sketch files can be generated from the React codebase. This means components in the Airbnb React repo, which already have product-level fidelity, can be populated with real data and rendered to a Sketch file. It’s another reward for those adventurous designers willing to learn React!
It’s also a touchstone technology, pointing us toward understanding mockups not as the source of truth, but instead as another target for automated output. With these generated files, we get a clear picture of which components are sitting in the repo. Best of all, a product designer relying on Sketch does not have to change their workflow and can use more accurate files (generated by the codebase, not by hand) to compose their work. At last, we can have confidence in how the components look at product-level fidelity and in how these components behave with real data.
React Sketch.app is great because it syncs the React and React Native repos and Airbnb’s design system Sketch files. But what about the Android and iOS implementations? How can designers make sure these are in sync? React Sketch.app points the way, but we’ll have to go further.
Here, we can learn from tools and WYSIWYGs of the recent past like Dreamweaver and Interface Builder. Tools in this category allow users to combine elements of UI, hook the UI up to data and interaction, and then export deployable code. Unfortunately, these software produce code that’s not maintainable by humans, and so few companies use them as an official part of their process. Did these tools promise too much and deliver too little?
Luckily, design system components are simple views with minimal logic. (Learn more about views and MVC architecture here and here.) Unlike the promise of tools like Dreamweaver and Interface Builder, components are easily exported view files that developer partners can incorporate into existing workflows.
Solving 2 problems with 1 tool
To create a designer-controlled source of truth, we’ll need next-generation design system tools to enable the composition of components and to automate the output of the design system to any number of target clients (codebases, vector files, and documentation sites).
As outlined in the section above, if we standardize an export format that contains production-level fidelity for components, then other interpreter modules can be freely built to compile a single component file into all the various flavors of production code. Then, varied implementations of any given component would flow from a single source of truth and would plug right into existing workflows — saving resource-intensive and error-prone human interpretation of mocked UI. Product designers and design system designers would finally control the UI they have always been responsible for.
At Airbnb, we’re tackling this challenge in a variety of ways, including with Lona, our tool for defining design systems and using them to generate cross-platform UI code, Sketch files, images, and other artifacts. A highly experimental prototype, Lona represents our exploratory approach to discovering what’s next for design at scale. Taking a research-based approach to the future of design systems encourages experimentation and collaboration.
Intelligent systems compose themselves
If you’ve not heard of Alan Kay, make sure to look him up. He’s credited with inventing the graphical user interface, Object Oriented Programming, and — with his concept of the Dynabook — even the tablet. Given that design systems fit firmly inside of the world that Kay and his peers built, it’s worth listening to him.
Once the magic of design system standards enables both private and open source development for cross-platform use, design systems will most likely reach new heights of functionality and popularity. Whereas siloed development requires talented people to reinvent the design system anew for each employer, standards-based design systems might easily plug into a passionate community that will add capabilities to our nascent tools.
Taking inspiration from Kay and the PARC maxim — “The best way to predict the future is to invent it” — I’ll sketch out a stretch goal for the future of design systems and tools in the paragraphs that follow.
Virtuous cycle — plugging into the feedback loop
Often, our design systems contain components that aren’t tracked in any special way. Without special consideration and considerable development effort, the design systems team is blind to the usage statistics and performance metrics directly associated with the system. To gain insight into design systems, teams must manually track which components each product team uses, and then try to glean usability information from product team UX research and performance metrics.
At Airbnb, we see data from the large proportion of our native apps, which now use design system components. Connecting these stats to user metrics lets us know which components might be underperforming and deserve special attention.
Similar metrics also have an internal benefit. Dashboards containing usage data aid the internal perception of the system among product designers and engineers alike. It’s easy to imagine that tracking the use of a design system can go beyond recording end-user impressions (views) and interactions. Even analyzing the usage of documentation, design files, and other internal tools can lead to insights that better enable product teams.
But why should such insights be limited to just a few companies with the resources to create similar tooling? A module could be built for just this purpose. It would track where canon components are used in product and then apply a flag to that component. These flagged components could be polled, and usage stats could come straight back to the design systems team. The insights would then be used to improve internal tools and the system itself.
Another promising direction would be for our components to become somewhat layout aware, and then declare their intended use to both the system at large as well as sibling components. Given this new super power, design system components could now communicate where they should generally be used and then share information about what types of data they generally contain. Even a simplified implementation of this functionality would help designers instantly swap abstracted data sets to, for instance, see how translation might affect the layout of a particular component given a longer language like German or a right-to-left language like Arabic.
This awareness might even go a step further with a given component stating the kinds of screens or interactions preceded by its own display, and again which kinds of interactions or screens should follow upon user interaction. This awareness would enable predictive assembly.
Predictive assembly — a pattern tool
Because each component now sends messages about where it’s typically used and what type of data it usually contains, we should also be able to see existing components predictively assembled into screens. A future tool might allow the product designer to view a particular component, click a randomize button, and see the component in situ with other components arranged in an order that more or less semantically makes sense.
By allowing designers to up-vote particularly useful component groupings (or patterns), a layout aware tool featuring predictive assembly — perhaps better described as a design system pattern tool — could become an instrumental way in which product designers quickly discover repeatable patterns in the system. With just a single click, a designer would be able to compose an effective solution made of many components and export the basic design for any project. Moreover, with the advent of standardization, design systems could become plug-and-play with compositions composed by predictive assembly, allowing for a quick way to assess the integrity of a given design system.
Predictive assembly would enable the product manager, developer, or product designer to provide a set of data and ask for a menu of pre-assembled screens that display the data. We could then choose the best option, only correcting the layout if needed.
It’s worth noting that if we get particularly good at predictive assembly and consult the right minds to take it further, we will soon find ourselves in a world where machine learning provides artificial intelligence for plug-and-play design systems. Among many other things, we might witness the birth of a world where phone users could eliminate branded apps, preferring a locally installed design system that could spin up UI without needing to load templates over the network.
As designers, we continue to rely on intuition, performance metrics, and user research to create products. If a large percentage of our users can complete a task and proceed to the next stage in the flow, then we generally call the design successful. We use templating languages to add options to a design so even more users reach their goals. But by connecting design systems to artificial intelligence, we could achieve a step change in customizable UI. Instead of designing products that work for most users, systems design and AI will apply specific solutions for specific individuals.
As digital product designers, we’re asked to ride the waves of change in our industry. People just like us helped small development teams create apps for desktop computing in the 1980s. In the late 1990s, other designers formed a coalition that pressured large software companies to unite front-end web development through the web standards project. In the mid-2000s, we focused on communication and user-generated content while developing patterns for creating web apps, and just a little later, all our work was available in mobile form factors like the iPhone. Now, after nearly 4 decades of designed computing, design systems and AI are shaking hands — presenting us with new opportunities for innovation at scale.
Every designer will now be asked how they might change custom, one-off design solutions into reusable components that grow the system. We’ll have to think more holistically, and work collectively to develop and learn a new set of tools.
How can we work together to improve our workflows and better scale design? What should be in a design system standard, and what kinds of tools could be built if such a standard existed? How will design systems and AI combine to create extremely customized interactions for end users? There are many questions, and each answer points to a possible future. What I love about design is that it enables each of us to explore the ideas we feel may have the biggest impact, and project our ideas outward and into the future. Innovation is certain.