At a time when AI is being utilised for getting answers and automating tasks, there is one thing that even the sophisticated systems like OpenAI’s Codex cannot do: challenge human creativity and visual taste when creating good design. On this, Andrew Ambrosino, the head of OpenAI’s Codex, the powerful coding application that CEO Sam Altman once famously admitted “made him feel a little useless”, said that creative professionals can breathe a temporary sigh of relief.While speaking on a recent episode of Lenny’s Podcast, Ambrosino revealed that jobs centered around design, artistry and aesthetics are much safer from immediate AI replacement than traditional technical roles. The reason, he explains, comes down to a fundamental limitation in how AI learns: design is incredibly hard to measure.
Why design defeats the machines
Ambrosino explained that teaching an AI how to design a beautiful user interface or a compelling piece of art is vastly more complicated than teaching it how to write computer code. With software engineering, an AI has an objective feedback loop: it can instantly test if a piece of code works simply by checking if it compiles and runs without crashing. Design has no such rules.“I think design’s a little bit harder to grade than software. In that, creating a loop where you can train the model on what’s good design and what’s bad design is just a little bit more tedious and onerous than, you know, does the code compile?,” Ambrosino saidBecause there is no simple algorithmic formula for “good taste,” AI struggles to independently judge whether its visual creations actually look good to a human eye. This technical limitation is why current creative AI tools often function well as productivity assistants, but completely fail to deliver a polished, finished product without human intervention.“It’s just a little bit more tedious and onerous than you know. Does the code compile does it? Do what it’s supposed to? Because the human aspect of taste is part of the feedback mechanism you need. I also think that through the labs historically and invest in making their models good at things that accelerates AI research. And that in the era, the early era of coding models, is very clear that the model being able to write correct code. Would accelerate research right in a way that you can’t really make the same case for design. Not that getting good at design isn’t important, it’s that it’s not directly in that that flywheel,” he said.
Navigating the era of ‘Vibecoding’
The current technological barrier means that instead of facing immediate extinction, creative professionals are shifting toward a collaborative relationship with AI. Designers, musicians and writers are adapting by learning new hybrid skills like engineering precise prompts or ‘vibecoding’, which is a term used to describe the practice of using natural language to dictate the overall aesthetic or “vibe” of a digital project while letting the AI handle the routine technical lifting.“Those are practical reasons. These models will get pretty good at design. There’s some kind of murkier things that is is going to be really tough. I have kind of a short list of them. One is there is an aspect of of culture to what is considered good design,” he said.
