OpenAI’s head of Codex, Andrew Ambrosino, said AI still can’t reliably handle creative design because design quality is far harder to measure than code, telling Lenny’s Podcast that training a model on what counts as good versus bad design is more tedious than checking whether code compiles. The gap comes down to taste and the judgment to know when something actually works, qualities that resist the clean grading loops AI uses to improve at programming. Figma CEO Dylan Field has made a similar argument, contending AI models trained on the distribution of data tend to produce designs people recognize as merely average rather than polished. Ambrosino closed by crediting human judgment, saying to give it up for the human brain for now.
