Making your design system AI-ready means clearing out “system debt” like hard-coded values and undocumented choices. The proposed fix treats design decisions as infrastructure in three layers. First, you store specs as structured Markdown files that the AI pulls on every generation. Second, you wrap design tokens in a closed set of named variables so the model can’t invent new values. Third, you run an audit script to flag any hard-coded overrides and a sync routine to catch stale specs before they drift out of date. Loading states breed UX friction when you scatter them across components. A better approach shifts data fetching to the router or application level. Preload route data aggressively, cache it locally, and handle transitions with a single global fallback. That way, users navigate into fully rendered pages instead of watching dozens of mini-loaders flicker on and off. Elsewhere in dev news: Homebrew 6.0.0 ships “tap trust,” forcing explicit approval for third-party repos, flips on its faster JSON API by default, and brings sandboxing to Linux. A security researcher uncovered an AMD AutoUpdate flaw: metadata over HTTPS but executable downloads via HTTP with no signature checks. AMD resisted the patch at first but eventually rolled out a fix. After WWDC 2026, Apple’s Foundation Models run on-device or via Private Cloud Compute, accept image inputs, support custom skills, and let you invoke server models through a unified Swift API.