Stripe's Katie Dill on Scaling Design with AI

Right after World War II, a massive building boom, the largest in history, was driven by new technologies and a push for rapid construction.

Modernism, initially founded on intentionality, became mass-copied without regard for context, leading to "zombie buildings" that lacked differentiation and care for user experience.

Today, AI tools enable small teams to execute projects that once required extensive resources, sparking an "AI building boom."

Stripe's Katie Dill cautions that this era echoes the post-war building boom, carrying similar risks of scaling without intentionality and creating monotonous, characterless software, or 'zombie UI'.

Large Language Models (LLMs) excel at providing the most probable answers, often at the expense of original, context-specific solutions.

AI's speed fosters a "temptation of done," where rapid execution can mask poor problem-solving and insufficient differentiation.

The ease of generating work with AI can lead to a "disposable" mindset, neglecting the long-term maintenance and user impact required for quality products.

For software to feel cared for and possess personality, it requires deliberate, small details.

Examples of such intentional design include the Grok animation that reacts to user input, the specific date on a calendar tab, and context-aware agents like Link, which demonstrate a builder's investment in the user experience and brand character.

Without a clear human point of view, AI will default to generic, backward-looking outputs that lack distinctiveness.

Stripe addresses this by defining its brand through optimism, which informs every design detail, and recognizes that distributed ownership makes alignment on design standards more crucial than ever.

The aim is to move beyond simply accepting AI's first drafts and to prevent the abdication of human judgment in the design process.

Stripe faced internal discussions regarding whether work generated by AI should be held to different quality standards than human-made work.

The company concluded that users prioritize the quality of the product itself, not the method of its creation, making 'meticulous craft' essential to improve outcomes beyond initial attempts.

Stripe uses the term 'Pepsi bubbling' to describe going beyond what customers might explicitly notice in design, emphasizing craft that delights even in subtle details.

Developing good taste in design requires noticing what users truly need and want, rather than solely relying on their stated requests.

It involves critically evaluating the world to discern what signals quality from mediocrity, constantly refining one's own standards.

Drawing inspiration from diverse fields like art and science beyond one's immediate product area helps to elevate and scale design standards.

Designers now face a landscape where agents and generative UI proliferate, rendering their direct presence in every decision impossible, as stated by the source.

This shift means the designer's role has evolved from crafting individual screens to building scalable design systems that can guide AI.

Just as Gutenberg's printing press enabled the creation of extensible yet opinionated systems for text, modern design systems must balance automation with craftsmanship to ensure beauty in the final product.

Stripe aims to empower builders to move from prompt to production almost instantly, leveraging AI for rapid development.

Initial attempts with a central design documentation system using AI for consistency proved unsuccessful, as identical prompts often produced inconsistent outputs.

Stripe's new CLI-based system integrates full templates and workflows, rather than just atomic components, ensuring a more coherent product.

This system is intentionally 'opinionated,' guiding AI toward expected product behaviors and embedding design standards directly into the means of production.

Traditional development processes inherently included quality filters due to staffing limitations, which restricted the volume of ideas that could be built.

Generative AI, however, allows for the creation of 20 ideas a week, shifting the critical quality control to a post-build editing phase.

The role of an editor becomes paramount in ensuring a cohesive, end-to-end user journey that transcends mere functionality.

Builders must break the ingrained habit of equating the completion of a build with its quality or finality.

According to Nabil Koreshi, great art is characterized by unexpected details and deeper underlying themes that AI often struggles to replicate.

AI frequently produces 'slop,' where details, such as color choices, appear arbitrary and lack intentionality, highlighting a gap in its creative capabilities.

Editors must assume accountability for every pixel, ensuring that each detail serves a purpose and contributes to the overall design intent.

An example from Stripe involved a team animation that required 56 iterations to achieve the desired realism and fluidity, demonstrating the extensive human oversight needed.

Rather than serving as a tool for monotony, AI acts as a creative catalyst that allows designers to explore more complex viewpoints.

It acts as a creative catalyst, fostering originality rather than leading to monotonous outcomes.

In an era where product creation is increasingly accessible, differentiation through unique and artistic design becomes critical.

Current interfaces, such as chats and command-line interfaces (CLIs), do not represent the zenith of modern interaction design.

Designers should strive to invent new interfaces and aesthetics, aiming for generative, alive, and dynamic systems.

AI presents an unparalleled opportunity to innovate new interaction standards and aesthetics, similar to the transformative impact of multi-touch technology or synthesizers on their respective fields.

For the 'Built to Grow' cover, Stripe utilized AI to refine a marble design initially created by a human artist.

AI was used to fine-tune colors and lines, ensuring that the original intent of the human artist was preserved and scaled.

The project's goal was to augment and scale human judgment, not to replace it with automated processes.

To avoid the 'burrito dilemma' and ensure unique results, users should incorporate brand direction and personal beliefs into their initial inputs.

Avoiding the "burrito dilemma" means pushing beyond the initial AI draft through critical review and iteration, and using adversarial agents to critique and refine the generated work.

Leaders must foster an environment that encourages exploration, rather than merely mandating AI usage.

Creating unique solutions, though more challenging, yields greater value than adhering to conventional, "cookie-cutter" patterns.

AI should serve to elevate the overall quality ceiling of products and experiences, rather than simply accelerating existing workflows.

John Ruskin's observations on Gothic architecture underscore the value of individual, non-identical details as expressions of human care and thought.

Users are more deeply impressed by thoughtful problem-solving and character-driven design than by the speed of production.

Designers have the opportunity in this digital building boom to avoid creating the digital equivalent of 'zombie buildings' by imbuing products with the 'hand and care' of the maker, fostering a creative renaissance.