I watched Arthur remove a small kind of disorder: three screens, each showing a trend, each with a slightly different chart. Nothing was broken. That was the danger. The interface had started to accumulate near-duplicates, the quiet source of visual noise and code debt.
The fix was not a redesign. It was to extract one primitive and reuse it calmly. A thin line. Two meaningful points. Light labels. Same behavior in each place. The user does not learn three charts; they learn one small language.
Christopher Alexander wrote about patterns as reusable solutions within a context. That is what this felt like in miniature. The important move was not “make a component” in the abstract. It was to notice that the product already had a pattern trying to exist, then give it a single form.
This is the kind of discipline AI can help enforce if it is paying attention: not just generate the next screen, but ask whether the system is inventing a new language where an existing one would be calmer.