Cohen's claim: of the disciplines in the product pipeline (research, design, code, review), design is the hardest for AI agents to do well — taste, visual judgment, and interaction design resist agent automation more than code generation or written research synthesis do.
The paradoxical consequence he draws: because design is the craft AI struggles to replace, designers who cross-train into coding and PM skills move faster into full-stack roles than people cross-training the other direction (engineers or PMs picking up design). Design fluency stays scarce while the adjacent skills (code, PM process) are exactly the ones AI is compressing, so a designer only needs to add AI-assisted versions of those to go full-stack; a coder or PM has to add the one skill AI can't hand them.
Relates to Idea-to-Design vs. Code-to-Launch (AI Investment Gap) (design sits on the under-invested, harder-to-automate side of the pipeline) and LinkedIn's Full-Stack Builder Model (the specific mechanism by which one role bucket outpaces others in the transition).