The nature of craft has shifted from production to judgment, especially with the rise of AI. Historically, design craft was demonstrated through polished artifacts like pixel-perfect files or functional prototypes. However, these artifacts never directly shipped revenue; they merely promised a product. True design impact is measured by user engagement, retention, and revenue.The advent of desktop publishing in the 1980s marked a similar transition, where those who adapted to the changing medium thrived. AI is accelerating this shift in design, making the production layer, like wireframes and initial drafts, increasingly commoditized and free. What remains valuable is the ability to choose the right problem and own its outcome.Craft now resides in discerning which of the many plausible options deserves to exist, a skill that AI, focused on efficiency, often overlooks. This requires moving from making to curating, where judgment and taste are paramount. Polish is now a baseline expectation, not a differentiator.This new literacy demands the ability to articulate craft into words; tacit knowledge is insufficient for machines. Translating intuitive understanding into explicit standards, constraints, and definitions of "good" is crucial. This process of writing down one's craft sharpens the underlying skill.AI can apply this articulated craft, holding standards and preventing anti-patterns, but only if the standards are clearly defined. Providing this explicit context, or "scaffolding," through formats like DESIGN.md helps AI agents operate within specific constraints rather than generating generic output. This standing context, established by humans, is the most valuable component.The focus is shifting from machines to people, with continuous discovery and iterative feedback from real users becoming more important than ever. AI can generate plausible research documents and personas, but the habit of checking these against actual people provides cheap insurance against misinformation. Iteration is now about rapidly killing weak options with evidence, not early commitment. Machine capabilities raise the value of human habits like research and iteration, preventing the dissemination of fictional findings.