Where's the Shovelware? Why AI Coding Claims Don't Add Up (www.learnui.design)

After 2.5 years of hype, the author argues there’s no evidence that current AI has sparked a design renaissance: no clear productivity boom, no mass layoffs of designers, and little speed-up in real-world workflows. Instead, gains are mostly in narrow, task-specific tools (background removal, content generation, layer renaming) and in enabling technically savvy creators to launch small apps or prototypes with less developer friction. Anecdotes of big wins coexist with frequent slogging and failures, and macro signals (flat app releases/domains/GitHub activity) don’t show a new wave of creative output. One-off chats with LLMs are a poor substitute for the iterative, contextual design process that produces high-quality outcomes. Technically, this traces to LLMs being prediction machines trained on median internet data: they perform well on common patterns but hallucinate on rare, tightly-constrained, or novel problems. AI struggles with pixel-perfect work, clever logo concepts, high information density, complex interactions, tight constraints, and anything “out of the training data.” The practical implication: designers should double down on complexity, interdisciplinarity, novelty, and brand-driven craft—areas where AI’s architecture limits it—while using AI for low-complexity, internal, or prototype work. In short, tools lower the barrier, but vision and craft become the decisive differentiators.