Design knowledge is stranded. Your component principles, tone of voice, and definition of done exist in your head, in scattered docs, and nowhere your AI tools can reach. Harald Kirschner's Config 2026 talk names this as the core dysfunction: every AI session resets to zero because institutional taste is invisible to the model.
The fix is architectural, not behavioral. Kirschner proposes a shared knowledge layer that connects design context directly to AI workflows, covering specs, prototypes, and design systems. The argument is not about prompting better. It is about making context persistent and structured so the AI stops being a amnesiac collaborator.
The talk is worth reading in full for how it frames the problem before the solution. The specific mechanism for building that knowledge layer, which tools connect to it, and what a working implementation looks like inside Figma are the details the description withholds. Those are the parts that determine whether this is philosophy or a deployable workflow.
[WATCH ON YOUTUBE →]