In this conference talk, Stripe's Katie Dill argues that AI's speed can produce generic, prematurely finished software. She proposes four ways to preserve care and originality: define a point of view, encode it into production tools, evaluate complete user experiences, and invest in creative exploration. Stripe's design tooling and creative projects illustrate the approach.
Topic breakdown
- The AI building boom and three quality risks · 0:06–5:27
- Define a point of view and a shared quality bar · 5:27–9:06
- Encode intent into design systems and tools · 9:06–12:10
- Edit the whole experience beyond done · 12:10–16:26
- Use AI for creative exploration and artistry · 16:26–21:48
The verbatim transcript is included for reference. Seven numbered figures accompany the topic notes, with explanatory captions below each image. Click an image to open the exact captured moment in the video.
Key takeaways
- Dill identifies three risks of AI building: generic answers, polished output that feels finished too early, and disposable work without long-term ownership.
2:172:443:56 - Quality should depend on the result users experience, regardless of whether AI made it. One Stripe ad critique ended with 17 specific improvements.
7:127:42 - Stripe's initial documentation-aware MCP produced inconsistent results. Its subsequent design-system CLI supplies documentation during building and includes full templates and flows to guide product behavior.
11:0911:2711:45 - Faster building shifts quality filtering toward post-build editing. Someone must judge whether the complete journey solves the user's problem and feels coherent.
13:0613:2313:45 - The event's AI-assisted opening animation took 56 iterations. Dill credits AI with allowing a more complex scene and more viewpoints while human scrutiny improved the result.
15:46 - Leaders should spend some of AI's production savings on exploration. Dill's final quality test is solving user problems with thoughtful details and brand character.
19:3221:16