Key Takeaways

  • OpenAI's Andrew Ambrosino, a product and engineering lead for the Codex app, states that design's subjectivity, driven by human taste, makes it significantly harder for AI to "grade" and learn compared to the binary success of code compilation.
  • AI research prioritizes capabilities like code generation because they directly accelerate the AI development flywheel, pushing design to a secondary focus.
  • Unlike software engineering, which often favors known patterns, good design typically requires a strong element of novelty and cultural context that current models struggle to generate.
  • A key barrier for AI is mastering the complex abstraction layer that connects a design's visual elements to its underlying software architecture and interaction patterns.

Why Taste Trips Up AI: The Ungradeable Truth

Ever wonder why AI can churn out code, but still can't nail a truly good design? Lenny Rachitsky posed this exact question to Andrew Ambrosino from OpenAI, and the answer isn't about complexity; it's about taste. Ambrosino, who helps lead the Codex app, points out a core truth: design is hard to grade. Software either compiles or it doesn't. It meets its spec, or it fails. This provides a clear, objective feedback loop for AI to learn from.

But design? That’s different. Ambrosino explains, “I think design's a little bit harder to grade um than than software and that you know creating a loop where you can train the model on like what's good design and what's bad design is just a little bit more tedious and ownorous than you know does the code compile does it you know do do what it's supposed to right it because the human aspect of taste is is like part of the feedback mechanism you need.” Good design isn't just about functionality; it's about how it feels, how it resonates, and that's a moving target defined by subjective human preference, not a compiler error.

AI's Own Flywheel: Why Design Got Skipped

The struggle with design also comes down to AI's internal incentives. The frontier of AI research often focuses on a self-perpetuating cycle: develop capabilities that, in turn, accelerate further AI development. Code generation fits perfectly here. An AI that can write accurate code speeds up the creation of more advanced AI. Design, while important, doesn't offer the same immediate boost to the core research flywheel.

Ambrosino clarifies, “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.” This means the incentives, historically, have pushed development towards code mastery rather than aesthetic brilliance. It’s a pragmatic choice, prioritizing the engine over the paint job for now.

The Novelty Trap and Abstraction Gap

Beyond taste and research priorities, AI faces two other hurdles: novelty and abstraction. Software engineering often thrives on known, repeatable patterns. You want reliable, predictable code. Design, however, craves the unexpected, the fresh, the culturally relevant. Ambrosino notes, “There's an amount of like novelty that is more important in design than it actually is in software engineering. Like software engineering, you almost you almost want it to overindex unknown patterns, right? Whereas design it's like no, there's an element of randomness here and and novelty, right?”

Finally, there's a complex abstraction layer AI has yet to master. A human designer can envision a user interaction pattern and translate it into specific visual elements, knowing the underlying code architecture. AI still struggles with this high-level semantic connection between an abstract concept and its concrete visual output. Ambrosino explains it as the difficulty in understanding “the semantics between these two things that look different like they're both in list that have the like this style that convey this interaction pattern to the user and I think like that is still feeling a little bit out out of reach with the current technology right that abstraction layer.” This gap means AI can't reliably bridge the chasm between design intent and implementation.

What to Do With This

If you're building design tools or a design-focused AI product, stop trying to automate taste. Instead, focus on building tools that augment human designers by handling the repetitive, pattern-based work, freeing them for the novelty and subjective curation AI can't yet touch. Specifically, identify three areas in your design workflow that are rule-based or data-entry heavy, and explore how AI could handle those, leaving the creative, context-dependent decisions to your human team.