How to scale intent, quality, and artistry with Al | Katie Dill (Stripe)

Summary

Katie Dill argues that the AI building boom risks producing “zombie interfaces”: polished-looking but generic products that ignore users, context, and brand. She compares this to postwar construction, when modernist styles were widely copied without the original movement’s intent.

AI tends to generate likely, familiar answers and makes creating things feel fast and disposable. Teams therefore need to supply the judgment and care that the tools lack.

AI should scale a product’s intent, quality, and creativity—not just its output or consistency.

Four Recommendations

1. Establish a clear point of view

Define what the brand stands for, who its users are, and what those users need. These standards matter even more when product decisions are distributed across teams and agents.

At Stripe, optimism informs details ranging from color and writing to the products themselves. In one advertising review, the team debated whether AI-made work should have a different quality bar. They concluded that users care whether the result is good, not how it was made. The review led to 17 refinements.

2. Encode standards into the system

When interfaces can be built without a designer present—or generated dynamically—design systems must communicate more than visual components. They need to capture product patterns, flows, behavior, and context.

Dill says Stripe moved beyond an MCP that referenced design documentation, which produced inconsistent results, toward a CLI built on its design system. The goal is to give developers more reliable, context-aware ways to move from a request toward production. The system should scale intent, not merely consistency.

3. Don’t confuse “ready” with “good”

AI makes it easy to generate many options, but shifts quality filtering to later in the process, when rejecting work can be harder. Teams need editors who assess the whole experience from a user’s perspective:

• Does it solve the problem?
• Does it fit how users think?
• Does it feel coherent?

Dill describes an event animation that took 56 AI-assisted iterations to reach a polished result. The example illustrates that AI can open creative possibilities, but still requires careful review and refinement.

4. Use AI to expand creativity

AI can encourage sameness, but it can also make new kinds of interactions and aesthetics possible. Dill urges teams to explore beyond familiar interfaces and protect unusual ideas.

To get better creative results, teams should provide specific prompts and source material grounded in the brand’s point of view, keep refining beyond the first draft, and use agents to critique the work. They should use some of the time and cost saved by AI to make something distinctive, rather than only doing existing work faster.

Closing Idea

Dill contrasts generic, mass-produced design with Gothic architecture, where individual details reflected the maker’s care. Users, she says, are impressed not by technical flourishes alone, but by products that solve their problems and show that the team anticipated their needs.

Main Speaker and Source

Katie Dill, Stripe.