Katie Dill on Design Quality in the AI Building Boom: Four Rules Against Zombie UI

Key points

1. 1

   Pick a point of view before the model picks one

   Models return the most probable answer: what already worked, or what is in style. Dill's example is a Korean barbecue site that reads like a software company. Her first rule is to decide who you are for your users and what they care about. That becomes your standard. When a team and its agents all build, ownership blurs and the call drifts to the model.

   “you have to have a point of view. If you don't, AI will give it for you.”

2. 2

   Same bar for AI-made work

   In a Stripe design review of an ad, a partner asked what the quality bar is for work made with AI. Dill's answer: users only care whether it is good. The review ended with 17 fixes, and the team now uses "Pepsi bubbling" as a verb for that level of care. She aims one level deeper than the customer can see, and she says perfection is not the goal.

   “It doesn't matter to the users how it was made. It matters to them if it's good or not.”

3. 3

   Design the system, and make it opinionated

   Screens now ship with no designer in the room, so the design system has to make the calls a designer used to make. Stripe first put an MCP server over its design docs. Three people typed the same prompt and got three different results. Stripe moved to a command-line tool built on the design system that serves the right doc at the right step and ships whole flows. Her model is Gutenberg's 290 characters, cut so machine type could feel handmade.

   “the old system scaled consistency, but the new system needs to scale intent.”

4. 4

   Refuse to confuse done with good

   Teams used to have 20 ideas and staff one, so pruning happened before the build. Now a team can build all 20 in a week, and the filter lands after. Dill wants a named editor who uses the work as a customer would and pushes half-formed work to a finish. Her team's opening animation for the event took 56 rounds with AI before it looked right.

   “the filtering that used to be throughout the process now needs to happen post build when it is a lot harder to say no”

5. 5

   Protect the strange

   Dill calls AI the greatest creative catalyst we have had. She says chat windows, charts and command lines are far from the best interfaces possible. When everyone can build, standing out gets harder. So she wants prompts fed with your own source material, and rival agents to critique the output. She asks team leads to give people room to explore.

   “Protect the strange. AI lowers the cost to create. Let's spend some of that savings on making something truly special.”