“Agentic design systems” bundles two separate ideas—using agents to do work and letting them make final calls—into one. The author builds agent-driven design tools daily and welcomes machine-legible systems. His gripe is with any setup that hands off the judgment layer—deciding what counts as a primary action, when to break the grid, or whether a component even belongs—to an autonomous agent loop without a named human in charge. He points out that the term quietly erases human accountability, even though most proponents, like Into Design Systems and Vicente Garcia, explicitly say some judgments always need a person’s stamp. He breaks it down further by comparing agentic loops to “vibe coding.” Both produce a component from a prompt, but vibe coding stops there. A design system must enforce standards over time. True agentic design systems follow generation with hard gates—token checks, linting, accessibility tests, design-parity reviews—each gate owned by a person. Without that, you get drift masked as official output. Brad Frost calls this “vibe engineering,” pairing automated generation with human-owned validation. The takeaway: agents should handle the mechanical 80 percent; humans must own the final 20 percent of judgment.