Siri, Clean the Dishes
Vibe Coding
Product Design
Sixty-three percent of people using AI to build interfaces in 2026 are not developers. They are product managers, founders, and marketing directors who discovered they could describe a screen and get something back that looked finished.
Sometimes it is. A well-prompted interface from someone who understands usability, hierarchy, and brand can cover a lot of ground. The problem is that most prompts don't come from people who understand those things. They come from people who know what they want the feature to do, and leave everything else to the model's defaults.
AI defaults are not good design. Without explicit direction - no design tokens, no component library, no documented visual language - the output converges on the same generic patterns. It functions. Users can navigate it. But there is a meaningful difference between an interface that works and one that is optimized for user experience. Flow logic, cognitive load, error handling, accessibility, the small decisions that reduce friction at scale - none of that is in the prompt, so none of it makes it into the output.
When that interface lands on a designer's desk for review, the work isn't editing. It's a reconstruction. Going through every screen praying you haven't missed a hidden feature you didn't build. Fixing usability issues nobody flagged because nobody was looking for them. Restructuring flows, resolving visual inconsistencies, replacing copy, reconnecting components to the system. Then engineering implements the revised version, which is effectively rebuilding the product from scratch.
The tool isn't the problem. A Roomba doesn't threaten a professional cleaner. It threatens the assumption that cleaning is a solved problem because the floor looks fine from a distance.
Vibe coding is genuinely useful in the right hands with the right inputs. A strong design system underneath it raises the floor dramatically - components, tokens, and documented constraints give the model something real to work with instead of reaching for defaults. Without that foundation, you're not moving faster. You're just moving the rework further down the calendar.